AM-49 · Prospecting UX research · 2026-10-09

What Kite should take from Clay, Autumn, Artisan and 11x about finding leads

I went through each product's public tours, docs, in-app recordings and API terms, then compared them with what appsmith-v2 ships today. All four have converged on the same experience: a chat that produces a live, cited list, a cheap preview before money is spent, and an explicit approval before anything is sent. Kite already has the chat, Slack, and a routed enrichment stack about as good as any of them. What it lacks is the list itself, the preview-and-approve step, and speed.

Clay: data platform + new agent Autumn: people-research agent + API Artisan: "Ava" AI BDR 11x: "Alice" digital worker No accounts created · no demos booked

Top recommendations

  1. Give every lead request a live list, not a CSV in task files. Rows should stream in next to the chat, with sources and verification on each cell, and Slack should get a summary plus the file. In Kite's own AM-10 eval, 6 of 8 "first 25 accounts" runs never got a list to the founder within the hour, and the two that did took 38 and 50 minutes. #1
  2. Show a sample and a cost estimate, then ask before spending. Clay ("nothing is spent until you approve"), Artisan (credits on every control) and 11x (criteria match rates + ETA) all do this. #2
  3. Run enrichment as one batched pipeline. Kite's baseline run fanned out into 24 agent sub-tasks over 52 minutes and cost $59.5. Clay runs columns over rows as code. #3
  4. Explain every row with a why-it-fits reason, criteria-as-columns, and which provider found each field. Kite already computes most of this and then buries it in the CSV. #4
  5. Make the ICP and signal watches saved objects: proposed persona cards, explicit qualification rules, and scheduled watches that add dated "why now" rows. #5 #6

Should Kite use their APIs?

Not as Kite-paid enrichment. All four terms forbid reselling the data or the service, and Autumn, Artisan and 11x also bar use in a competing product. Details

Clay
bring-your-own Public API + partner-OAuth MCP, on the customer's own credits. "not to re-sell any data".
Autumn
pilot w/ contract A clean async Task API with cited rows. "internal business purposes", no resale, thin privacy posture.
Artisan
connector at most No REST API. Inbound webhook plus an OAuth MCP into the customer's own org.
11x
no The API reference is unpublished, it is Enterprise-gated, and it has no data endpoints.

Keep Treg, Deepline and the contacts router as the paid path, and measure them (the capability matrix has 0 measured rows today).

Clay

Spreadsheet-first GTM data platform, now with a chat agent (Alpha, 2026-10-08)
  • Free filter preview, then approve the estimated cost
  • Waterfalls showing per-provider credit cost
  • "Why it fits" column, excludes owned accounts

Autumn

Chat research agent over a 1B-person index (YC W26, 2 people)
  • Chat on the left, a cited table streaming on the right
  • Each criterion becomes a MATCH column
  • "validation clean (0 failures)" receipt

Artisan

"Ava" AI BDR persona; self-serve from $0, $250 and $600/mo
  • Ava proposes persona cards with counts
  • Test a signal on 50 companies, with sources, free re-sample
  • Credit cost on every control

11x

"Alice" digital-worker SDR; sales-led, from ~$36–45k/yr
  • Sentence → scored criteria with match % and ETA
  • CRM-status column, continuous monitoring
  • Autonomy as settings: thresholds, escalation

Product teardowns

How each product goes from "I need leads" to a usable list. Screens are real product UI wherever the vendor made it public: in-app recordings, docs screenshots, click-through tours and product-video stills. Marketing renders are labelled. Click any screenshot to enlarge it.

Clay: from spreadsheet to agent

A GTM data and orchestration platform that started as a spreadsheet: Find People/Companies sources, enrichment columns, waterfalls across 150+ providers, and the Claygent AI column. On 2026-10-08 (Sculpt) Clay launched a chat agent called "Clay" (Alpha), plus a Knowledge Hub, alongside Audiences, Signals, Sequencer and Workflows. The UI evidence below comes mostly from frames of Clay's own in-app Loom recordings embedded in its docs. Sculpt recap · agent docs

From "I need leads" to an enriched audience

  1. Start from a sentence. Find leads → "Start with natural language: Sculptor will set up your filters". The text becomes an editable filter panel (cross-entity: companies × people × jobs, with exclusions), and a free live preview appears ("50 of 3,440"). "Filtering doesn't cost credits." docs
  2. Save as a table or an always-on Audience. An Audience "keeps feeding in new records as they match, automatically deduped". Results land in a draft, with new and existing records shown side by side. docs
  3. Or ask the agent. It states its assumptions ("Your saved ICP is…; 'recently raised' will mean the past 12 months"), asks what "best fit" means, and returns a ranked table with a Why it fits column and linked sources. Loom
  4. Refine and exclude what you own. "Exclude accounts we already own or have engaged" drops accounts with a CRM owner or recorded outreach. Each step leaves a typed artifact card (Search · 2 results · View).
  5. Stop before paid enrichment. "Email enrichment has not run… Should I enrich these two leaders' work emails?" is answered with a one-click Enrich and save. Bigger jobs get an approval card showing the estimated cost: Deny / Allow once. The promo shows "Enrich 312 matches? Estimated 118 credits… nothing is spent until you approve".
  6. Enrich through visible waterfalls. Providers run in order, each tile showing its credit cost, and the first hit wins. The work-email waterfall starts with free pattern inference, can add a validation provider, and offers Conservative/Balanced/Aggressive strategies and an optional "successful provider" column. docs
  7. Hand off as a draft. The agent drafts a Sequencer campaign: "Nothing was sent or enrolled". You press Launch yourself. Asked "why did 25 become 2?", it answers with a stage-by-stage drop table. Any task can be scheduled; a run that needs input pauses with "Waiting on you". docs

Quick read

Agent model
Not a persona: "Clay" is the product. There are three AI surfaces: the global agent (chat), Sculptor (copilot in tables) and Claygent (a per-row research column). A permission level sits in the composer: View only / Ask before runs and sends (default) / Allow all. The agent won't launch campaigns or create tables.
Data
Its own index ("900M people, 70M companies, 300M jobs"). A 150+ provider marketplace in waterfalls (Prospeo, Hunter, PDL, Apollo, Lusha, ContactOut…). Email status values: Valid / Invalid / Catch-all / Unknown / Role-based. Signals: job change, hiring, funding, news, web intent.
Trust cues
Per-provider credit chips, "~" estimates on AI columns ("try 5 rows first"), per-cell run status, "Worked for 3m 29s · Read 2 memories", artifact cards, an explanation of the funnel, and draft-first Audiences.
Pricing (2026-10-09)
Two meters since March 2026: Actions (orchestration, a few tenths of a cent) and Data Credits (from $0.05). Free; Launch $167/mo; Growth $446/mo. Third parties quote $185/$495 month-to-month. No result means no charge. The HTTP API, CRM auto-sync and webhooks need Growth+. pricing
API
Public API v0 (search, async routines, JSONL batches, credits, signed webhooks) on all plans. MCP at api.clay.com/v3/mcp with OAuth 2.1 and open DCR for partner products, capped at 20-row pages and 100 results. No numeric rate limits are published. The ToS says "You also agree not to re-sell any data you obtain from Clay." bring-your-own only
Complaints
Credit burn and a learning curve ("2,000 credits in the first week"). Key automation is gated to $446+/mo. The agent and Knowledge Hub are Alpha. Reported ratings: G2 ~4.7 vs Trustpilot ~2.2 (third-party).

What the screens show

Sentence first.
Sentence first. Find Companies: "Start with natural language… Sculptor will set up your filters", with suggested prompts.In-app Loom frame · source
Free preview before spend.
Free preview before spend. Editable cross-entity filters, "Preview 50 of 3,440" and a Sculptor chat to refine. A banner warns that 294 members can spend without a budget.In-app Loom frame · source
Autonomy in the composer.
Autonomy in the composer. View only / Ask before runs and sends / Allow all, with prompts suggested from the Knowledge Hub.In-app Loom frame · source
Assumptions → ranked list.
Assumptions → ranked list. The agent states its assumptions, asks what "best fit" means, and returns a ranked table with Why it fits and funding links. The Artifacts panel is on the right.In-app Loom frame · source
Exclude what you own.
Exclude what you own. A Search artifact card with a plain-English filter, after "exclude accounts we already own or have engaged".In-app Loom frame · source
Stop before paid enrichment.
Stop before paid enrichment. "Email enrichment has not run" plus a one-click "Enrich and save".In-app Loom frame · source
One batched question card.
One batched question card. Clarifying questions paged "1 of 4", each with a reason line and Skip.In-app Loom frame · source
Approval with estimated cost.
Approval with estimated cost. "Approval needed": estimated actions and credits per run, Deny / Allow once.In-app Loom frame · source
Review segment.
Review segment. Enriched emails, "+150 more" fields and suggested next steps.In-app Loom frame · source
Draft, not sent + "why 25 → 2".
Draft, not sent + "why 25 → 2". Sequencer draft with manual Launch, and the agent's stage-by-stage drop table.In-app Loom frame · source
Waterfall transparency.
Waterfall transparency. Provider order with a credit cost per provider. The winning provider is highlighted.Docs image of app · source
Cost estimate on AI columns.
Cost estimate on AI columns. "Try 5 rows first to estimate costs… 75% of runs cost less than 1 credit per row".Docs image of app · source
Cell-level progress.
Cell-level progress. "5% of cells completed" with success and remaining counts, plus auto-update and daily source runs.Docs image of app · source
Classic route: pick a source.
Classic route: pick a source. New table → Signals (track job changes, track new hires, webhook) or Companies (find companies, find your ICP from a domain, find jobs), each with a credit chip.Interactive tour step (recorded Nov 2024) · source
Preview before import.
Preview before import. Location filters and a limit, then "Previewing 50 of 1,000+ results, 100 of which will be imported".Interactive tour step · source
Pick enrichments, priced per row.
Pick enrichments, priced per row. "Would you like to add additional enrichments? Using providers with your own API key is free." Each option shows its credits per row. Auto-update and auto-dedupe are on.Interactive tour step · source
The spreadsheet.
The spreadsheet. The imported Find Companies table: one row per company, typed columns, "Add enrichment", and table history.Interactive tour step · source
Results live in cells.
Results live in cells. A "Find Contacts at Company" column shows "Found 12 People" or "✕ No Profile Found" per row.Interactive tour step (recorded Feb 2025) · source
Where credits went.
Where credits went. Settings → Credit usage, split by workspace or by integration (AI, People Data Labs, YouTube), with export.Interactive tour step · source
Sculpt promo.
Sculpt promo. "Enrich 312 matches? Estimated 118 credits… nothing is spent until you approve" and "41 of 312 scored ≥ 70" with a reason each.Official promo video frame (stylized) · source

Not captured: a Claygent column's output, the classic Find People panel, CRM sync screens and Signals setup. Clay's public "shared tables" now redirect to login, so these come from docs text and images only. The classic-table frames above come from Clay's own interactive tours, recorded in 2024–25, so the styling predates the 2026 agent UI.

Autumn: chat → cited table over a people index

Identity check. "Autumn.ai" is Autumn AI, Inc. (autumn.ai), a San Francisco company in YC W26 with 2 people: founders Vishnu Sampathkumar and Shiv Kampani. Its own docs say it is "a separate company from useautumn.com (billing software)". Other "Autumn"s are the Stripe billing tool (YC S25, London) and an employee-wellbeing app that joined Qualtrics. Confidence is high that this is the product the issue meant. docs · YC

"Everything you need to know about anyone." It claims an index of 1,060,831,012 people and 109,851,694 companies, plus a chat research agent ("Beanstalk 2.1") and an async Task API. The YC launch pitched it as "Stalk your prospects at scale", built on real-time signals. The real app needs sign-in, so the UI below comes from the scripted product mocks embedded on autumn.ai. The component names match the docs, but these are not live screens. site · sales prospecting

From "I need leads" to a cited list

  1. Type the ICP into a chat ("Find 25 Series A startups building AI dev tools…"). Saved tasks sit in a sidebar as named jobs, for example "Delaware Incorporations" and "Org structure for Greptile". A CSV of up to 10,000 rows can be attached for enrichment. docs
  2. Plan, then visible steps. The agent writes a plan "before it spends anything" and shows a tool timeline in plain verbs: "Scraped 18 funding announcements", "Saved ai-devtools-series-a.jsonl". It posts an interim note: "Saving the 18 qualifying rows now… then expanding toward 25." It asks at most one blocking question. docs
  3. Rows stream into a table beside the chat. Each cell carries a source-count badge, and a Sources tab holds the evidence. The API mirrors this: every value is {value, source_id}. docs
  4. Each criterion becomes a MATCH column, and the run ends with a receipt: "25 rows saved, all three rules match with sources on every row, validation clean (0 failures). Done."
  5. ICP fit + reasoning columns. A 3-dot fit score, per-rule MATCH pills, and free-text evidence such as "16 eng / 20 ops-reviewer of 36 classified" or "Hiring ML roles: no".
  6. Drill into profiles. Person profiles show identity, socials, email, education and experience. Company profiles have Overview / Financials / People (org tree) / Market / News tabs, plus a tech stack with a source count for each item.
  7. Refine in place, schedule, hand off. "Fixing a list should cost one message, not a rerun." Tasks can run in cron mode. Autumn calls itself "the research layer": CRM sync is Enterprise-only, and sequencing needs another tool. principles

Quick read

Agent model
A plain research assistant, not a persona. High autonomy by default ("plan, infer defaults, and execute without asking"). Three model tiers: ranger / scout / wren.
Data
"The open web: incorporation filings, LinkedIn, X, GitHub, event pages, news, code, and company sites." It uses a hybrid of indexed entities and live research. No email-verification vendor, waterfall or accuracy benchmark is published.
Trust cues
Source counts per cell, MATCH rule columns, a validation receipt, a tool timeline, and per-task credit usage plus queue ETA in the API.
Pricing (2026-10-09)
Hobby $20 (2K credits), Growth $75 (8K, adds person/company profiles), Scale $280 (32K, adds org trees and Slack bot), Enterprise custom (adds CRM). Its own article estimates ~$0.10 per shallow enrichment and up to ~$0.80 per deep one. Each task takes 2–10 minutes. pricing
API
The REST Task API (api.autumn.ai, OpenAPI published, X-API-Key) is on every plan. Starts are limited by credits; reads are capped at 180/min. No webhooks or MCP in the docs (claimed only on AWS Marketplace). The ToS limits use to "internal business purposes" and bans resale and "creating competitive applications". partial: needs a contract
Risks
No opt-out for indexed people. The CCPA table claims no identifiers were collected. The product surfaces personal email and high school. Its own mock shows scraping a Delaware registry whose banner prohibits automated mining. There are no independent reviews.

What the screens show

Chat entry.
Chat entry. Saved tasks in the sidebar and an ICP prompt in the composer (New task · Auto · Settings).Embedded product mock · source
Steps + streaming table.
Steps + streaming table. "Tool execution · 5 tools" with an interim plan note, while the table on the right fills (16 of 25 rows).Embedded product mock · source
Cited table + receipt.
Cited table + receipt. 25 rows with source counts in every cell: "validation clean (0 failures). Done."Embedded product mock · source
Criteria as columns.
Criteria as columns. Each prompt constraint is a MATCH column with its own sources.Embedded product mock · source
ICP fit score.
ICP fit score. "Agent ICP Targets": fit dots plus a MATCH column per ICP rule.Embedded product mock · source
Reasoning in the list.
Reasoning in the list. ICP notes, eng-vs-ops counts, ML hiring, founder background and a Sources column.Embedded product mock · source
Person profile.
Person profile. Identity grid (email, socials, school), about and experience.Embedded product mock · source
Company org tree.
Company org tree. Tech stack plus an org tree with department tags and LinkedIn links.Embedded product mock · source
Signals feed.
Signals feed. "Foliage" scan of funding and launch posts parsed into company, round and amount.Public interactive resource · source
Collection-method risk.
Collection-method risk. The agent scrapes Delaware ICIS while the page's banner prohibits automated mining.Embedded product mock · source
Identity from weak signals.
Identity from weak signals. Demo video, Delaware filing, investor clue, email pattern, GitHub org and LinkedIn role are resolved into one record, "and shows you where every claim came from".Public page section · source
API: rows with sources.
API: rows with sources. Output kinds (research/person/company). Every cell carries a source id.Docs page · source
Pricing and gating.
Pricing and gating. $20 / $75 / $280 credit plans. Profiles, org trees, Slack bot and CRM are gated by tier.Public page · source

Not seen: the real app at platform.autumn.ai (sign-in), the contents of the Sources tab, email-verification labels, and export, Slack and CRM screens. A free account (500 credits, no card) would show them; we did not sign up.

Artisan: "Ava, the AI BDR"

An outbound platform built around a named, face-bearing agent persona. Ava 2.0 (May 2026) is a ground-up rebuild that added autonomous replies, a dialer, a Signals page, "Chat with Ava" and credit-based self-serve pricing. Evidence comes from Artisan's own click-through tours embedded in the help center and from help-center app screenshots dated July–September 2026. Ava 2.0 post · help center

From "I need leads" to a contacted list

  1. Ava learns the company. She auto-builds a knowledge-base profile (positioning, products, proof, competitors) and flags each field "We guessed this" or "We couldn't find this". A Gaps tab turns questions she couldn't answer into tasks. docs
  2. Everything starts as a campaign with a type (cold, warm, upsell, website visitor, intent signal) and a data source: 250M+ B2B contacts, 200M+ local businesses, lists, CRM, CSV or webhook. docs
  3. The ICP becomes a list three ways. Ava proposes 3–5 persona cards with prospect counts. A plain-language search is turned into filters. A filter panel shows a live count ("100 of 768"). docs
  4. Enrich and research. Email and phone come from a waterfall over "22+ sources". AI web research adds a free-text question as a list column, with variables, an output type, a missing-input policy and a 5-credit "thinking mode". docs
  5. Signals trigger campaigns. A library of 14 signals, each card showing the positive-response-rate range seen across customers. A live preview shows "125 leads matched in the last 60 days". A custom AI signal is written in plain English, and "Test on 50 companies" returns Match/Unclear verdicts with sources, with a free re-sample. docs
  6. Qualify before contact. A "Let Ava research and qualify each lead" toggle adds "qualify only if / disqualify if" rules, plus an explicit policy for when no answer is found. Each check costs credits. docs
  7. Review and launch. "Require approval before sending" routes drafts to Tasks → Pending, where they can be approved or regenerated with instructions. Settings show the maximum 30-day credit cost and runway before launch. HubSpot/Salesforce sync dedupes on import, and a Slack notification matrix routes each event type. docs · Slack

Quick read

Agent model
A named persona (face, "she") in the sidebar with "Chat with Ava". Autonomy is a set of toggles: approval, autopilot replies, escalation, DNC. "Full self-driving" is marked Soon.
Data
"250M+ verified B2B contacts". Waterfall over 22+ sources, Bombora intent, website-visitor de-anonymization. Per-field provider attribution was not seen in real UI.
Trust cues
Credit cost on every control. Match/Unclear plus source chips on test runs. Contacted/Excluded/Pending per list. Disqualification reasons. CPL per campaign.
Pricing (2026-10-09)
Free $0 (300 credits/mo). Intern $250/mo annual (10K credits, ~500 leads/mo). Employee $600/mo (25K, ~1,250 leads/mo). Email 2 credits, phone 10, ~20 credits per contacted person end to end. Annual, non-refundable. pricing
API
No public REST API. A per-campaign inbound webhook (10 rpm, Employee+) and an OAuth MCP at api-dashboard.artisan.co/mcp (30 tools, on the customer's own account). The ToS bans redistributing contact data and competing use. not a provider
Complaints
Trustpilot 4.1/42: credits burn fast, generic AI copy, very low reply rates, annual lock-in. LinkedIn was dropped after the Jan 2026 LinkedIn ban. Trustpilot · TechCrunch

What the screens show

Auto-built company profile.
Auto-built company profile. Each field is flagged "We guessed this" or "We couldn't find this", with an onboarding checklist (3/11).Help-center app screenshot · source
Ava proposes the ICP.
Ava proposes the ICP. "Ava suggests these personas for you to target": editable cards with filter chips and ~prospect counts.Interactive tour step · source
Find leads.
Find leads. A "Describe who you want to find" box plus filters, a live result count (768), then Save & add to campaign.Interactive tour step · source
Pick a data source.
Pick a data source. B2B 250M+, local 200M+, lists, CRM, CSV, webhook. Credit balance and "cost per lead contacted ~9" in the header.Help-center app screenshot · source
AI web research column.
AI web research column. A free-text question with variables, output type, missing-variable policy and "Thinking mode" at 5 credits per contact.Help-center app screenshot · source
Research lands as a column.
Research lands as a column. "Most important sponsored event in 2026" across a 28,315-person list. Hover shows the full answer.Help-center app screenshot · source
Signal library with benchmarks.
Signal library with benchmarks. 14 signals, each with a positive-response-rate range measured across customers.Interactive tour step · source
Funding signal, live preview.
Funding signal, live preview. Round, amount and 60-day look-back filters. The right pane previews matched leads before launch.Interactive tour step · source
Test before you spend.
Test before you spend. A custom signal run on 50 companies: Match/Unclear per lead with a reason and source chips. "Re-run on a new sample: No charge".Interactive tour step · source
Qualification rules.
Qualification rules. "Let Ava research and qualify each lead" (2 credits per check), plus what to do when she can't find an answer.Help-center app screenshot · source
Approval and guardrails.
Approval and guardrails. Require approval before sending, autopilot replies, senders and exclusions.Help-center app screenshot · source
Slack routing matrix.
Slack routing matrix. Each event type (escalated replies, meetings, visitor alerts) goes to email, a Slack channel or a DM, at a set frequency.Interactive tour step · source
Self-serve credits.
Self-serve credits. Free / Intern $250 / Employee $600 per month (annual), with per-action credit costs.Public page · source

Not seen without an account: the main lead-search page, the lead profile drawer (including whether provider attribution is shown), the approval queue, the Chat with Ava UI and analytics. A free plan exists (300 credits, no card), but we did not sign up.

11x: "Alice", the digital-worker SDR

Named AI workers for B2B revenue teams. Alice does outbound (email, LinkedIn, consented calls, SMS, WhatsApp), and Julian handles inbound voice and chat. 11x is sales-led on annual contracts with no self-serve trial, and its September 2026 video is titled "Introducing 11x: The AI Growth company". Evidence comes from the public docs (86 pages), stills from 11x's "Launch Sequence" product videos, and UI renders on 11x.ai, which are labelled mockup below. docs · Launch Sequence

From "I need leads" to booked meetings

  1. Admin setup takes 2–4 weeks: sending domains, knowledge base, ICP, sequence, review. The CRM (Salesforce, HubSpot or Pipedrive) must be connected first. Managed Gmail inboxes come with domain purchase, warm-up and rotation. docs
  2. The knowledge base is scoped per campaign and versioned with rollback. It includes "boundaries", meaning claims Alice must never make. docs
  3. Pick a play from a gallery tagged with the sources each play accepts and needs: event registration, cold outbound, website-visitor retargeting, lost opportunities, inactive leads. docs
  4. Build the audience. Filters exclude CRM-owned accounts, open opportunities and customers. Live Web Search turns a sentence into numbered criteria, each with a live match rate, then shows profiles analysed, preview contacts, a confidence score and an ETA. You can also start from a CRM list with a Qualification Agent prompt set to "Continuous Monitoring". docs
  5. Campaign Strategist (mid-2026) is chat-first. It asks analyst questions, drafts the persona, positioning and research instructions, and "shows its work… every research step and every source, visible in a live timeline". launch page
  6. Deep research per lead produces a report with cited insights and a recommended angle, steered by directives such as "research 10-K for Gen AI initiatives". The docs say to qualify before spending because "Alice is priced per prospect". docs
  7. Approve, then send. There are three modes: review each message, an approval workflow, or autopilot, and the docs warn "Don't start on autopilot". Smart Replies can be tested with a simulator. Results write back to the CRM, and Slack alerts tag the lead owner. docs

Quick read

Agent model
A named persona ("Hi, I'm Alice"). Rebuilt in 2025 from a 5-step wizard into a chat-first supervisor with sub-agents on LangGraph. LangChain talk
Data
"400M+ verified contacts", described both as "built into the platform" and as "21+ premium data providers". A mockup credits People Data Labs. No per-field waterfall or confidence is shown publicly. docs
Trust cues
Criteria match rates, confidence and ETA before launch. Cited research. An "Action Required" queue. KB version history. Domain and mailbox health. A published "honest ramp curve".
Pricing (2026-10-09)
Alice Growth from $3,750/mo billed annually (2,000 prospects/mo). The FAQ on the same page says "starts at $36,000 per year". Pro and Enterprise are custom. Billing is per prospect, not per send. pricing
API
Webhooks and an API are described, but "The endpoint reference is not published yet". They are gated to Enterprise (docs) or Pro+ (pricing table). There are no search, enrich or find-email endpoints, and no MCP. The ToS bans resale and building competing products. not a provider
Credibility
In 2025, TechCrunch reported logos of non-customers and ARR counted inside break clauses. 11x said the logos were "human error" and cited 79% retention. The founder-CEO stepped down in May 2025. TechCrunch

What the screens show

Sentence → scored criteria.
Sentence → scored criteria. Live Web Search splits the ask into 3 criteria with match % each, plus profiles analysed (102), preview contacts, confidence and "Estimate Time: 20 mins".Product video still · source
Qualify a CRM list.
Qualify a CRM list. A Qualification Agent prompt over a CRM list, "Continuous Monitoring" sourcing and a preview table.Product video still · source
Play gallery.
Play gallery. Plays tagged by required source (CSV, CRM, Website). The sidebar shows Action Required / Hot Leads.Product video still · source
Knowledge scoped per campaign.
Knowledge scoped per campaign. Play Research → pick all or specific KB resources, each with a processing status.Product video still · source
Chat builder + live table.
Chat builder + live table. Chat on the left; accounts with a CRM-status column and leads with live research-status chips on the right.Marketing mockup of UI · source
Lead drawer.
Lead drawer. "Lead Verified", ICP facts labelled with the data provider (People Data Labs), and agentic research with source logos.Marketing mockup of UI · source
Autonomy as settings.
Autonomy as settings. Tone, approval thresholds ("Deals > $10,000"), escalation rules, send windows and channel mix.Marketing mockup of UI · source
Simulate before going live.
Simulate before going live. Reply categories plus a Test-scenario simulator.Product video still · source
Enrichment as value.
Enrichment as value. Contacts and accounts enriched per month and verified emails enriched (112,357), framed as "their dollar value".Product video still · source
API status.
API status. "The endpoint reference is not published yet".Docs page · source

Not seen: the live app, because there is no self-serve signup and seeing it needs a sales contract. Interactive tours don't exist on 11x.ai. YouTube blocked video downloads from this host, so the frames above are YouTube's own auto-stills.

Side by side

The same 14 questions for each product, with Kite today in the last column. Product claims are as published on 2026-10-09; sources are linked in each product section above and listed at the bottom.

Clay Autumn Artisan 11x Kite today
Shape of productData + orchestration platform for GTM engineers and RevOpsPeople and company research agent plus API ("the research layer")AI BDR: list, enrich, sequence, reply, dialAI SDR "digital worker", sold as a managed outbound programmeAI CMO in chat and Slack; lead lists are one research job among many
How you startA sentence becomes editable filters, or you ask the agent. Templates and a lookalike-from-domain option exist.Type the ICP into a chat, or attach a CSVCreate a campaign, pick a type and data source; Ava suggests personasPick a play from a gallery tagged by source; 2–4 weeks of admin setup firstAsk in chat or Slack; the CMO delegates a research task
ICP inputKnowledge Hub ICP and personas (agent proposes edits) plus a filter panel with exclusionsNatural language; each constraint becomes a rule columnAuto-built company KB (guessed / couldn't find), persona cards with counts, filter panelKB scoped per campaign; sentence → numbered criteria with match % eachInferred from wiki icp/ prose or the request; no editable object
List surfaceTable or Audience (always-on, deduped, draft first); typed artifact cards in the chatSplit view: chat left, table streaming right; Sources tabLists inside campaigns; Contacted / Excluded / Pending countsAudience preview, then campaign; the CRM is the system of recordA table in the reply plus contacts.csv in task files (500-row read-only preview)
Enrichment visibilitywaterfall shown provider order, credits per provider, winning provider column, email status valuessources per cell no provider or verification vendor named"22+ sources" waterfall no per-field provider attribution seenhidden "21+ providers"; only a mockup names PDLin the CSV email_source and status per row, plus routes tried; not rendered
Signals / intentJob change, hiring, funding, news, topic intent, web intent (Growth+)Formation (Delaware), funding and launch feed, GitHub orgs14 signals incl. Bombora intent, website visitors, custom AI signals with a test runJob changes, funding, hires, tech shifts, website visitors to the personPredictLeads news, jobs and financing (signal-first lane); ads activity
Qualification"Why it fits", score thresholds, Claygent columns, funnel explanationFit dots, MATCH per rule, reasoning columns with counts"Qualify only if" rules + policy for unknowns; Match / Unclear with sourcesQualification Agent before spend; disqualification reasons to review1–5 fit score with a one-line reason; gates in cost order (prose)
Review & approvalPermission level per task; approval card with estimated cost; drafts never auto-launchAt most one clarifying question; review via "continue" on the same task"Require approval before sending" queue; max 30-day credit cost shown before launchReview each / named approver / autopilot; "Don't start on autopilot"; reply simulatorChat questions; no spend approval; prose drafts for human-sent channels
Export / CRMHubSpot/Salesforce actions, Sheets, ads audiences; auto-sync is Growth+CRM sync Enterprise-only; "send to your CRM/sequencer"HubSpot (Intern+), Salesforce (Enterprise); dedupe on import; CRM lists as DNCBi-directional Salesforce/HubSpot/Pipedrive sync with research written backOnly when the task names a store, via a generic connector; no dedupe
Outreach hand-offSequencer draft; the user presses LaunchNone (by design)Email sequences, dialer, autonomous replies; LinkedIn dropped after the Jan 2026 banEmail, LinkedIn, phone, SMS, WhatsApp; managed inboxesResend single sends; LinkedIn and calls drafted for a human
Agent presentation"Clay" agent (chat), Sculptor (copilot), Claygent (column); no personaPlain assistant; model name shown ("Beanstalk 2.1")Named persona with a face: "Ava… she"; "Chat with Ava"Named persona: "Hi, I'm Alice"; chat-first Campaign Strategist"Kite", a named CMO with specialists
Progress & trust cuesWorked-for timer, memories read, cell-level run status, artifact panel, Alpha caveatTool timeline in verbs, interim plan note, source counts, validation receiptCredits on every control, list counts, CPL, "takes up to 10 minutes" expectationCriteria match %, confidence, ETA, "Action Required", research-status chips (mockup)Hand-off receipt only; no working state by design (2026-09-30)
Pricing modelActions + Data Credits; Launch $167, Growth $446/mo; no result means no chargeCredits: $20 / $75 / $280 per month; ~$0.10–0.80 per enrichmentCredits: $0 / $250 / $600 per month (annual); ~20 credits per contacted personPer prospect: from $3,750/mo annual (2,000 prospects)Kite credits (LLM + provider spend); no per-list estimate
API for KiteBYO onlypilot w/ contractconnector at mostnon/a

Which experience fits a chat- and Slack-first agent?

Fits Kite: list builders that grew a chat (Clay agent, Autumn)

Both treat the chat as the steering wheel and the table as the deliverable. Each step leaves an artifact you can open, and every cell carries evidence. That is exactly the gap between Kite's reply and Kite's CSV. Clay's agent adds the guardrails Kite lacks: a permission level, cost approval, "exclude what we already own", and drafts that never auto-launch. Strength: transparency and control. Weakness: the user still has to read a table. Clay is complex and credit-hungry, and Autumn is young, with a weak privacy posture.

Partly fits: AI SDR personas (Artisan Ava, 11x Alice)

The persona is marketing. Kite is already a named agent and gains nothing from a second face. What is worth copying is how the persona is bounded: persona cards the user approves, tests on a sample before scaling, approval queues, escalation rules, send windows, and Slack routing by event. Strength: outcome framing (meetings, CPL) and onboarding that learns the company. Weakness: lock-in contracts, credit burn, generic copy and low reply rates in reviews, plus 11x's 2025 credibility problems. Their sequencing and mailbox infrastructure is out of scope for Kite.

Kite today: the baseline

Read from appsmith-v2 origin/main at 8b9f24689e6 (2026-10-09). Paths link to the file on GitHub. Real-run numbers come from the AM-10 comparison eval (deploy previews, 2026-10-01).

How a user gets leads in Kite

  1. Ask in chat or Slack. The CMO agent turns the request into a brief and delegates a Research task. The brief always carries a "people outcome" (execs or speakers, LinkedIn + email). (prospect-research/SKILL.md)
  2. ICP is inferred from the team wiki (icp/, positioning/) or from the request. If neither has it, Kite asks. There is no ICP form and no saved filter set.
  3. Discovery is web research (Exa, Parallel, page reads) that produces candidate rows: name, title, company, domain, LinkedIn, source URL. A signal-first lane pulls PredictLeads funding, hiring and news events and drops anything older than about 90 days.
  4. Qualification gates run in cost order (identity, then cheap filters, then evidence scans). Each row gets a 1–5 fit score with a one-line reason. Large lists fan out into batch and per-account sub-tasks.
  5. Emails come from the routed contacts tool native:contacts-find-work-emails: Treg, then Deepline batches, then Apollo as a fallback. Every address is verified, personal mailboxes are excluded, and nothing is guessed. (contacts-router.md)
  6. Delivery is a table in the task result plus a contacts.csv with a fixed 12-column header. The web Artifacts page shows that CSV read-only, first 500 rows. CRM loading happens only when the task names a store, through a generic connector call.
  7. Progress: background tasks deliberately show no working state (decision 2026-09-30, #20954). The hand-off reply acts as a receipt, and the list arrives later as a message.

What a real run looked like

Case outbound_signal_meetings: "Help us book 20 sales meetings with B2B marketing teams… start with the first 25 accounts." 8 runs: 4 prompt/model arms × 2. Each run had a 60-minute budget on a fresh team. AM-10 report

Arm · runTasksBilled $CSV rowsReached the founder?
baseline · 12559.525no, CSV stayed in task files
baseline · 21622.715no
opus · 12528.30no
opus · 235.30no
cmo-prompt · 13447.50no
cmo-prompt · 21120.80no
all-prompts · 11326.626yes, in 38 min
all-prompts · 2920.625yes, in 50 min

In baseline run 1, all 25 rows had an email. 13 were verified, 11 were "valid-risky; unverified" and 1 was invalid ("do not email"). 22 of the 25 came from the routed contacts tool. The root task ran 52 minutes and fanned out into 24 sub-tasks, about one per account. Caveats: 2 runs per arm, 5 concurrent runs per preview, and billed cost includes LLM plus provider spend. The data is in ~/kite-compare-eval/runs/*/outbound_signal_meetings.

Data providers wired today (people, companies, signals)

SourceHow it's reachedWhat it gives KiteCost / balance model
Treg routed6 native tools (tool_gateway_treg.py)Company search/enrich, people search/enrich, find work email, verify email. Each call routes across a provider pool (26 children for company enrich).Per call, depends on provider. Company enrich floor is $0.0018. No balance endpoint.
Deepline routedPlatform catalog + onboarding clientA "unified GTM data API" over about 100 providers. Batch email workflows of up to 100 contacts. Reverse-email lookup.Per workflow run, not measured.
Contacts routernative:contacts-find-work-emails1–1,000 rows per call: Treg sync, Deepline batches, Apollo fallback, Treg verify on every address. Runs as a 24h Redis background run.Capped by max_cost_usd (default $0.25 per person).
Apollo via MonidPlatform catalogPeople search with masked emails, plus a people/match reveal.One prepaid Monid wallet. A single 402 disables Apollo, LinkedIn, X, Akta and Context.dev together.
Crustdata4 native tools + 21-tool hosted MCP1B+ person profiles, company search/enrich, LinkedIn posts, work email, job search, technographics.0.03 credit per search row; 1–7 credits per person enrich.
PredictLeads6 native toolsSignals: news (37 categories), job openings (27), financing, similar companies.1 credit per company; discovery billed per row.
Exa · ParallelNative, behind researchSemantic people/company discovery, event rosters, page reads, FindAll entity lists.Per query; FindAll pro is $10 + $1 per match.
LinkedIn via MonidCatalog (scrape)Profiles, employees, posts, jobs. Rate-limited, and not used as a list-building source.Monid wallet.
Clay MCPnot wiredRemoved in #18676 (badc6bc990d, 2026-09-15). The roster lists it as evaluating: "Card exists; no registry door. Re-wiring is a separate decision."Clay credits; the MCP reports balance as booleans only.

Sources: ROSTER.md, CAPABILITY-MATRIX.md (46 capability claims across 14 sources, 0 measured: hit rate, precision and cost per hit are all still "—"), clay-mcp.md, atlas README.

Gaps the repo already names

No list object

Research state lives in the result text. provider-runs.json holds no rows, and the wake guard finds "delivered contacts" by scanning prose for addresses. Design question Q1 (a typed ledger per task) is still open. design-questions.md

No people/company entities

Kite has no dedupe, no suppression against the CRM, and no "already contacted" check. Q7 (canonical person and company entities) is open. The only "Leads" page in the app lists form submissions from Kite-built websites (frontend/src/data/Leads/types.ts).

Outbound stops at the CSV

The atlas lists sequences, suppression lists, engagement webhooks and "a person-safe lead export to CRM/ESP connectors" as needs wiring. Email goes out as single Resend sends. A prospect list reaches a CRM only when the task names one (prospect-research Output). Inbound leads are logged to a connected CRM (email-campaigns), but outbound lists get no CRM dedupe or sync.

What they add over Kite's current providers

Kite's paid path is the routed contacts tool (Treg, then Deepline, then Apollo via Monid, with every address verified), plus Crustdata, PredictLeads, Exa and Parallel. Coverage figures below are vendor claims. None of the four publishes accuracy that could be compared with Kite's, and Kite's own capability matrix has no measured hit rates yet. Where a product's real advantage is workflow rather than data, the table says so.

CapabilityKite today Clay Autumn Artisan 11x
B2B people & company coverageCrustdata 1B+ profiles; Apollo's database; Treg routes company enrich across 26 providers; Deepline fronts ~100overlaps Own index of 900M people / 70M companies plus 150+ providers. Many are ones Kite reaches already (Apollo, PDL, Hunter).adds footprint 1.06B people / 110M companies from the open web: X, GitHub, registries, events, relationships, org treesadds SMB 250M+ B2B plus 200M+ Google Maps local businessesnothing visible "400M+ verified"; providers unnamed
Work email find + verifyRouted batch of 1–1,000 rows, verification on every address, catch-all and risky flags kept, personal mailboxes excluded, nothing guessedmore providers per row ~10 named email providers in one waterfall, a validation step and risk strategies. Claims to "routinely triple" coverage (unverified).no evidence Emails on request; no verification vendor namedsimilar Waterfall over 22+ sources, "verified" badge, bounce test before sendopaque Verification stated as a platform guarantee
Phone numbersExcluded by policy (prospect-research enrichment rule 4)yes Phone waterfall (PDL, ContactOut, Selligence)on requestyes 10 credits per number, plus a dialeryes Phone, SMS, dialer
Buying signalsPredictLeads: news (37 categories), jobs (27), financing, similar companies; ads activity. No job-change tracking.adds Job change, promotion, new hire, topic intent, web intentadds Company-formation (Delaware) and a funding and launch feed parsed from X/LinkedInadds Bombora topic intent, champion tracking, website-visitor de-anonymization, custom AI signalsadds Job changes, competitor reviews, renewal windows, person-level website visitors
Per-row research at a known priceAgent research per candidate inside sandbox tasks; slow, and costed only after the factworkflow Claygent / AI columns with a "~" estimate and "try 5 rows first"workflow Each prompt rule becomes a MATCH column with sourcesworkflow AI web-research column at 5 credits; qualification at 1–2 credits per checkworkflow Deep research per lead with directives; per-prospect pricing
CRM-aware dedupe and exclusionNone (open design question Q7)yes Excludes owned or engaged accounts; Audiences dedupeEnterpriseyes Dedupe on import; CRM lists as DNCyes CRM-status column; owned accounts excluded
Keeps the list freshCron Workflows exist, but there is no persistent list to append toyes Always-on Audiences; daily source runsyes Tasks in cron modeyes Signal cadence, e.g. monthly at 1 credit per companyyes "Continuous Monitoring" sourcing
Published accuracy evidenceCapability matrix: 46 claims, 0 measuredsome Publishes provider data testsnonenonenone
Net: on core B2B contact data, the four add little that Kite's routed stack can't reach. Clay's email waterfall is the only credible "more hits per row" claim, and it can't be resold. The real gaps are signals Kite doesn't have (job-change and champion tracking, website-visitor identification, third-party topic intent, company formation), phones (a policy choice, not a missing provider), and above all workflow: priced per-row research, CRM-aware dedupe, and lists that stay fresh. Before adding any vendor, check whether Deepline's ~100 fronted providers or Treg's pool already cover job changes and visitor identification.

APIs, webhooks, MCP: can Kite use them?

Checked against each vendor's official docs and terms on 2026-10-09. "Provider" means Kite calling the API with its own account to serve its customers. "Bring-your-own" means each customer connects their own account and pays the vendor directly.

SurfacesAuth · gating · limitsPricingTerms that matterVerdict for Kite
Clay Public API v0 api.clay.com/public/v0: query-mode people/company search, async routines/{id}/run (1–100 items, 202 → complete), JSONL batch runs, credits/balance, signed (HMAC) completion webhooks. MCP api.clay.com/v3/mcp: search (free), functions, Audiences (Enterprise). CLI and agent plugin. developers.clay.com · MCP docs API key per user and workspace; API on all plans (HTTP API integrations, auto-sync and webhooks in-app are Growth+). MCP: OAuth 2.1 + PKCE with open DCR "for a partner product… on behalf of a signed-in Clay user"; the admin must allow unknown clients; DCR is limited to 4 burst / 10 per hour per IP. No numeric rate limits are published (429 + Retry-After). MCP pages are 20 rows, capped at 100 results, with per-user monthly credit caps that hard-block. rate limits · MCP FAQ Same credits and actions as in-product, "no additional cost". Search filtering is free. Data Credits from $0.05. ToS §4: "You also agree not to re-sell any data you obtain from Clay." Credits can't be transferred without approval. No separate API terms found. ToS bring-your-own only
Not as Kite-paid enrichment. A customer who already pays for Clay could connect their workspace. Use the routines batch path with webhooks or patient polling for more than 20 rows, not MCP. The docs explain Kite's 2026-08-07 incident: MCP is built for one rep's research, not a fleet of concurrent tasks.
Autumn Task API api.autumn.ai (OpenAPI published, 22 routes): POST /task (prompt) or /task/start (spec with output.schema, target_count), SSE streams, partial rows readable mid-run, continue/stop, /credits, CSV upload. Output kinds: research, person, company. No webhooks and no MCP in the docs (both claimed only on AWS Marketplace). openapi.json · docs X-API-Key or Bearer; API on every plan. Starts are limited by credit balance; GET /task/{id} is limited to 180/min, 20k/day and 2 in flight; one turn at a time per task (409). rate limits Credits ~$0.0088–0.01 each; ~$0.10 per shallow and ~$0.80 per deep enrichment; 2–10 minutes per task. §1: licence "for your internal business purposes". §8: may not "resell… distribute, or transfer rights… Prohibited activities include creating competitive applications". No published DPA, subprocessor list or opt-out for indexed people. ToS · privacy pilot with a contract
Technically the best fit of the four. An async task with a schema and cited rows maps straight onto Kite's research tasks. Only consider it as a deep-research lane (org trees, formation, technographics, non-LinkedIn footprint) after a written embedding agreement and a DPA, with personal fields stripped.
Artisan No public REST API (api.artisan.co/v1 answers "Token not found"; no developer docs). Inbound webhook per campaign. MCP api-dashboard.artisan.co/mcp ("early feature", 30 tools: search, enrich with a credit quote, lists, export, campaign drafts; can't launch). webhook docs · MCP docs Webhook: a per-campaign sk_live_ key, 10 requests/min per campaign, Employee plan and up. MCP: OAuth to the customer's Artisan org (DCR and PKCE seen in discovery), all plans. No outbound event webhooks. Credits: email 2, phone 10, ~20 per contacted person. May not "distribute, sell, or license… contact data made available by Artisan… to third parties", nor "create or compile… a database", nor use the service "in connection with any commercial endeavors that compete with our Services". ToS not a provider
At most a user-connected integration, so customers who already run Artisan could push Kite-built lists into their campaigns. Low priority.
11x API + webhooks are described (push data in; pull outreach, research and outcomes out; trigger Alice/Julian), but "The endpoint reference is not published yet". No search, enrich or find-email endpoints. No MCP or SDKs. webhook docs Auth and limits unpublished ("contact your 11x representative"). Gated to Enterprise for Alice (docs), though the pricing table ticks Pro and Enterprise. billing FAQ Bundled in the tier; from $3,750/mo billed annually. §2.1 "internal business purposes"; §2.2.4 no "developing a similar or competing product"; §2.2.7 no "rent, resell or otherwise allow any third-party access". ToS no
The API runs 11x's own outreach and doesn't return lead data. Use 11x as a UX reference only.

Recommendations for appsmith-v2

Ranked by impact for customers who use Kite to find leads, then by effort. Effort is rough: S ≈ 1 engineer-week, M ≈ 2–4 weeks, L ≈ 1–2 months. Each item names the competitor screens it comes from and the parts of the Kite repo it touches. None of them changes Kite's chat/Slack-first model. They give the chat a list to point at.

1

Give every lead request a live list, not a CSV in task files

Autumn and Clay's new agent both put the chat on the left and a table on the right that fills row by row, with a source behind every cell. In Kite the list is prose plus a contacts.csv attached to a task. In the AM-10 eval, 6 of 8 "first 25 accounts" runs never got a list to the founder within the hour. One of them built a 25-row CSV the founder never saw. Repo design question Q1 (a typed per-task ledger) already describes the backend half.

Seen in: Autumn chat + streaming table; Clay agent artifacts panel and audience review; 11x "preview contacts start to pop up"

What to build / where: A typed list ledger written through a validated CLI (Q1), the prospect-research Output contract, a new list view next to pages/Artifacts and pages/Tasks (replacing the 500-row read-only TextArtifactPreview for lists), and a Slack summary with the CSV attached via slack_io.upload_file. The wake guard can then read rows instead of scanning prose (utils/contact_addresses.py). A list is a deliverable the user opens, not a busy indicator, so this stays consistent with the 2026-09-30 no-working-state decision.

P0effort Limpact 5.0/5
2

Show a sample and a cost estimate, then ask before spending

Clay's agent ends each plan with an approval step that shows the estimated credit cost, and Sculpt says nothing is spent until you approve. Clay columns suggest running 5 rows first. Artisan prices AI qualification at 2 credits per check, right on the toggle. Kite runs spent $5–$60 each with no estimate up front. The contacts router already has max_cost_usd, but the user never sees it.

Seen in: Clay approval with estimated cost, "try 5 rows first"; Artisan credits on the qualification toggle

What to build / where: The first research turn returns 5–10 qualified sample rows, the criteria it applied, an estimated population, and estimated credits and time. The CMO turns that into an approval card (reusing the CmoGoalApprovalCard / CmoProposalCards pattern) and Slack buttons. Paid enrichment runs only after a "go". Touches the CMO receipt rule, work-delegation, the contacts router budget, and Billing's usage breakdown.

P0effort Mimpact 4.8/5
3

Run enrichment as one batched pipeline instead of an agent fan-out

Clay runs columns over rows deterministically and shows per-cell progress ("5% of cells completed"), with optional daily auto-update. Kite's baseline run fanned out into 24 per-account agent tasks over 52 minutes and cost $59.5. The repo's open Q9 asks which stages should be code. Search, enrich and verify are code-shaped; fit reasons and why-now are model-shaped.

Seen in: Clay run progress and cell status, waterfall; Autumn parallel agent steps

What to build / where: A list-run capability: discovery returns candidate rows, then code runs the enrichment and verification columns in batch over the ledger. native:contacts-find-work-emails already takes 1–1,000 rows, and the Treg and Crustdata batch endpoints can do the same. Model calls are spent only on judgment columns. Touches work-delegation fan-out rules, contact_email_service.py, and research-enrichment-program/PLAN.md Layer B.

P0effort Limpact 4.6/5
4

Explain every row: why it fits, evidence links, and where each field came from

Clay's agent ranks results with a "Why it fits" column. Autumn adds reasoning and source columns next to the ICP score, and every cell links to evidence. Clay's waterfall shows which provider tried each field and what it cost. Kite already computes fit_reason, email_source, email_status and the routes it tried, but they end up as CSV text.

Seen in: Clay ranked list with "why it fits", waterfall providers; Autumn ICP notes, reasoning and sources; Artisan AI signal test run with citations

What to build / where: Cell-level provenance chips and verification badges in the list view (#1), plus a "routes tried" tooltip from the contacts router output. Small once #1 exists.

P1effort Simpact 4.0/5
5

Make the ICP a saved, editable object with qualification rules

Artisan builds a company knowledge base from the website, suggests personas, and turns qualification into rules ("qualify only if…", plus what to do when no answer is found). 11x's Alice learns product, positioning and proof points from a knowledge base. Autumn scores against explicit ICP rule columns. Kite re-derives the ICP from wiki prose on every request.

Seen in: Artisan KB autoprofile, suggested personas, AI qualification rules; Autumn ICP rules; 11x knowledge base

What to build / where: A structured icp/ wiki page schema (firmographic filters, personas, rules, exclusions, signals), proposed during onboarding from the research Kite already runs (onboarding_research), editable on the Knowledge page, and reused by every list and watch. Touches wiki-management, prospect-research Inputs, and pages/Knowledge.

P1effort Mimpact 4.0/5
6

Signal watches that add net-new leads on a schedule

Artisan ships a library of intent signals (funding, hiring, champion job change, website visitors, custom AI signals), each with a live preview. Autumn shows a daily signals feed. Clay sources can auto-update daily, and 11x sells plays such as website-visitor retargeting. Kite has a PredictLeads signal-first lane and cron Workflows, but nothing that packages them as "watch for this and add rows".

Seen in: Artisan intent-signal library, funding preview, custom AI signal; Autumn signals feed; 11x plays

What to build / where: A "lead watch" workflow template (manage-workflows) that takes the saved ICP and a signal and appends dated why-now rows to the same list, with a Slack digest. PredictLeads tools exist today; website-visitor identification would be a new provider.

P1effort Mimpact 3.8/5
7

Make lists CRM-aware: dedupe, suppress, and push approved rows

Clay's agent excludes accounts "we already own or have engaged". 11x and Artisan run on bi-directional CRM sync and import CRM lists to qualify. Kite has no person or company entity model (open Q7) and loads a CRM only when the task names one, so the same people can come back on every list.

Seen in: Clay company search excluding owned accounts; 11x CRM list + qualification agent; Artisan data sources incl. CRM

What to build / where: A per-team entity table keyed on domain and LinkedIn URL (Q7 option b), a suppression check against a connected HubSpot or Salesforce through Composio before delivery, and a "Send to CRM" action on the list. Touches tool-discovery-execution, connectors, and the list view.

P1effort Limpact 3.6/5
8

Hand off to outreach through an approval queue on the list

Artisan has a "request approval before sending" switch and a per-event Slack notification matrix. 11x exposes approval thresholds, escalation rules, channel mix and send windows as agent settings, and lets you simulate a reply before going live. Kite drafts per-contact messages as prose, and email goes out as single Resend sends.

Seen in: Artisan campaign approval settings, Slack notifications; 11x agent configuration, test scenario; Clay sequence draft

What to build / where: A draft column on the list with bulk approve, edit and send, and Slack approve buttons. Explicit per-team autonomy settings (approval threshold, channels, send window) belong in team Settings. Kite should not build a sequencer or mailbox warm-up; leave those to the team's own tool through connectors.

P2effort Mimpact 3.0/5
9

Pilot Autumn as a measured discovery provider; keep Clay as bring-your-own

None of the four can be Kite's paid enrichment: Clay, Autumn, Artisan and 11x all forbid reselling their data or service. Autumn is the one technically worth a measured pilot. Its async Task API returns cited rows against a schema, and it covers ground Kite's stack handles poorly: org trees, company formation, technographics, and footprints beyond LinkedIn. It needs a written embedding agreement and a DPA first, with personal fields stripped. Clay should be a bring-your-own connector for customers who already pay for it: their own workspace, the batch routines API (not MCP) for anything over 20 rows, and Retry-After honoured. That avoids the shared-credential 429 storm from 2026-08-07.

Seen in: See API verdicts

What to build / where: Run the existing Layer A source probes (backend/evals/source_probes/, task be:probe-sources) on a small approved budget before wiring a gateway door. If it wins, add it as a routed capability in tool_gateway_platform_integrations.py with capability declarations and prices, so it lands in CAPABILITY-MATRIX.md.

P2effort Simpact 3.0/5

Impact vs. effort

Hover or focus a dot for the item. One series; position carries the meaning.

3 4 5 S M L Effort → Impact → do first big bets 1 2 3 4 5 6 7 8 9

Impact is judged against the lead-finding job: list delivered, trusted, and used. Effort is a rough engineering estimate. Both are judgment calls, not measurements.

Sources

Every URL cited in the research notes behind this page, grouped by product, plus the links used inline above. All were visited between 2026-10-09 09:00 and 13:00 UTC.

Clay (65)

  1. https://www.clay.com/blog/sculpt-2026-recap
  2. https://developers.clay.com/index.md
  3. https://university.clay.com/docs/global-agent
  4. https://www.loom.com/share/fd24b393b15d495196740e457d2b4dfb
  5. https://university.clay.com/docs/search
  6. https://university.clay.com/docs/find-your-icp-based-on-your-domain
  7. https://www.loom.com/share/c6fca0b1ae974f03a81c47834066911a
  8. https://university.clay.com/docs/building-a-data-waterfall
  9. https://university.clay.com/docs/work-email-waterfall
  10. https://university.clay.com/docs/email-verification-statuses
  11. https://university.clay.com/docs/ai-pricing
  12. https://university.clay.com/docs/hubspot-integration-overview
  13. https://www.clay.com/pricing
  14. https://university.clay.com/docs/scheduled-tasks
  15. https://university.clay.com/docs/run-progress
  16. https://developers.clay.com/tables.md
  17. https://university.clay.com/docs/knowledge-hub
  18. https://university.clay.com/docs/actions-data-credits
  19. https://www.clay.com/waterfall-enrichment
  20. https://university.clay.com/docs/data-test-work-email-providers
  21. https://developers.clay.com/signals.md
  22. https://www.clay.com/changelog
  23. https://university.clay.com/docs/sculptor
  24. https://university.clay.com/docs/workflows
  25. https://blog.gojiberry.ai/blog/clay-pricing
  26. https://university.clay.com/docs/credit-budgets
  27. https://api.clay.com/public/v0
  28. https://developers.clay.com/llms.txt
  29. https://developers.clay.com/openapi.json
  30. https://developers.clay.com/public-api/authentication.md
  31. https://university.clay.com/docs/clay-api-cli
  32. https://developers.clay.com/routines/clay-managed-functions.md
  33. https://developers.clay.com/searches.md
  34. https://developers.clay.com/public-api/rate-limits.md
  35. https://developers.clay.com/public-api/webhooks.md
  36. https://university.clay.com/docs/webhook-integration-guide
  37. https://api.clay.com/v3/mcp
  38. https://university.clay.com/docs/connect-to-clay-mcp
  39. https://university.clay.com/docs/mcp-troubleshooting-and-faqs
  40. https://www.clay.com/mcp
  41. https://github.com/clay-run/agent-plugins
  42. https://www.clay.com/terms-of-service
  43. https://www.clay.com/dpa
  44. https://bouncezero.io/clay-io-review-2026
  45. https://www.g2.com/products/clay-com-clay/reviews_and_filters
  46. https://www.loom.com/share/9ff78557620b49b694f2f33dc658a7b5
  47. https://assets.clayrun.dev/recap%20scupt%20blog/1%20-%20Our%20pal%20Clay.mp4
  48. https://assets.clayrun.dev/MCP/ClayMcpFlow%20(3
  49. https://www.clay.com/claygent
  50. https://www.clay.com/workflows
  51. https://developers.clay.com/quickstart.md
  52. https://developers.clay.com/searches/advanced.md
  53. https://developers.clay.com/routines/api.md
  54. https://developers.clay.com/routines/batch-runs.md
  55. https://developers.clay.com/audiences.md
  56. https://developers.clay.com/use-cases/enrich-leads-and-accounts.md
  57. https://university.clay.com/docs/find-people-overview
  58. https://university.clay.com/docs/signals
  59. https://university.clay.com/docs/http-api-integration-overview
  60. https://university.clay.com/docs/send-table-data
  61. https://university.clay.com/docs/mcp-settings
  62. https://university.clay.com/docs/mcp-for-reps-vs-agent-plugin
  63. https://demo.arcade.software/eCSvICcORZsW7UAT61xE
  64. https://demo.arcade.software/HDAz4Q1nqDCWhxIUDGJz
  65. https://demo.arcade.software/tJuRZ1xjgrr73uBQAWEI

Autumn (52)

  1. https://www.autumn.ai/terms
  2. https://www.ycombinator.com/companies/autumn-ai
  3. https://www.autumn.ai/docs/llms.txt
  4. https://www.fastaijobs.com/companies/autumn
  5. https://ycombinator.com/companies/autumn
  6. https://www.cbinsights.com/company/autumn-2
  7. https://www.g2.com/products/getautumn/discuss
  8. https://www.ycombinator.com/launches/PNL-autumn-ai-stalk-your-prospects-at-scale
  9. https://youtu.be/v37JMhX8GVc
  10. https://www.autumn.ai/
  11. https://www.startuphub.ai/news/claudes-corner-autumn-ai-yc-w2026
  12. https://www.autumn.ai/pricing
  13. https://www.autumn.ai/customers/brex
  14. https://x.co…
  15. https://www.autumn.ai/verticals/sales-prospecting
  16. https://www.autumn.ai/docs/faq
  17. https://www.autumn.ai/docs/concepts/tasks
  18. https://www.autumn.ai/docs/guides/human-in-the-loop
  19. https://www.autumn.ai/#companies
  20. https://www.autumn.ai/resources/foliage?mode=scan
  21. https://www.autumn.ai/verticals/startup-data
  22. https://www.autumn.ai/docs/api/errors
  23. https://www.autumn.ai/docs/principles
  24. https://www.autumn.ai/llms.txt
  25. https://www.autumn.ai/resources/articles/best-people-search-apis
  26. https://aws.amazon.com/marketplace/pp/prodview-6tlf5oxf4ypfw
  27. https://www.autumn.ai/docs/concepts/outputs
  28. https://www.autumn.ai/docs/skill
  29. https://www.autumn.ai/docs/vibecoding
  30. https://api.autumn.ai/openapi.json
  31. https://www.autumn.ai/resources/articles/autumn-vs-exa-people-search
  32. https://www.autumn.ai/docs/guides/streaming
  33. https://www.autumn.ai/docs/concepts/lifecycle
  34. https://www.autumn.ai/docs/api/rate-limits
  35. https://www.autumn.ai/docs/concepts/model-tiers
  36. https://www.autumn.ai/docs/guides/accounts
  37. https://www.autumn.ai/resources/articles/people-search-api-gtm-engineering
  38. https://api.autumn.ai
  39. https://www.autumn.ai/docs/api-routes
  40. https://www.autumn.ai
  41. https://www.autumn.ai/skill.md
  42. https://www.autumn.ai/privacy
  43. https://www.autumn.ai/docs/guides/follow-up
  44. https://www.autumn.ai/#connections
  45. https://www.autumn.ai/#people
  46. https://www.autumn.ai/resources
  47. https://www.autumn.ai/resources/foliage
  48. https://www.autumn.ai/resources/founders
  49. https://www.autumn.ai/resources/founders/stealth
  50. https://www.autumn.ai/docs/llms-full.txt
  51. https://www.autumn.ai/docs
  52. https://www.autumn.ai/docs/cookbook/reviewed-list

Artisan (45)

  1. https://www.artisan.co/
  2. https://www.artisan.co/blog/artisan-launches-ava-2-0-first-autonomous-ai-bdr
  3. https://www.artisan.co/pricing
  4. https://support.artisan.co/knowledge-base
  5. https://support.artisan.co/getting-started/launch-first-campaign
  6. https://support.artisan.co/cold-outbound-campaign
  7. https://support.artisan.co/find-leads
  8. https://support.artisan.co/csv-as-a-data-source
  9. https://support.artisan.co/ai-web-discovery
  10. https://www.artisan.co/ai-lead-generation
  11. https://support.artisan.co/intro-to-signals
  12. https://support.artisan.co/configuring-a-signal-campaign
  13. https://support.artisan.co/targeting-page
  14. https://support.artisan.co/intro-to-website-visitors
  15. https://support.artisan.co/create-a-sequence
  16. https://support.artisan.co/build-messaging-for-your-campaign
  17. https://support.artisan.co/configure-campaign-settings
  18. https://support.artisan.co/what-happens-after-you-launch-a-campaign
  19. https://support.artisan.co/intro-to-crm
  20. https://support.artisan.co/configure-dnc
  21. https://support.artisan.co/slack
  22. https://support.artisan.co/claude-connector
  23. https://www.salesrobot.co/blogs/artisan-ai-review
  24. https://www.artisan.co/terms-of-use
  25. https://techcrunch.com/2026/01/07/yes-linkedin-banned-ai-agent-startup-artisan-but-now-its-back
  26. https://trust.artisan.co/
  27. https://support.artisan.co/intro-to-dialer
  28. https://yespress.io/artisan
  29. https://www.11x.ai/guides/artisan-pricing
  30. https://api-dashboard.artisan.co/mcp
  31. https://api.artisan.co/v1/webhooks/campaign/{campaign_id}
  32. https://support.artisan.co/webhook-documentation
  33. https://www.trustpilot.com/review/artisan.co
  34. https://deepline.com/gtm-stack/providers/artisan
  35. https://demo.arcade.software/j7T2yIIkl00ZxvP8aNqT
  36. https://demo.arcade.software/FAzzkjcV4yf5CopqpZfU
  37. https://demo.arcade.software/6opIP7M6u2gHWL1e2NRE
  38. https://www.artisan.co/ai-sales-agent
  39. https://artie.ai/
  40. https://support.artisan.co/llms.txt
  41. https://demo.arcade.software/zfrhxklKSLJOxH2u0KhY
  42. https://demo.arcade.software/iDIGxBDeJw7KvU69486f
  43. https://api-dashboard.artisan.co/.well-known/oauth-protected-resource
  44. https://auth.artisan.co/.well-known/oauth-authorization-server
  45. https://api.artisan.co/

11x (50)

  1. https://docs.11x.ai/introduction
  2. https://docs.11x.ai/alice/overview
  3. https://docs.11x.ai/help-center/overview/11x-pricing-overview
  4. https://www.11x.ai/
  5. https://www.youtube.com/feeds/videos.xml?channel_id=UCurs3NI5ss4EaNBnjLdiE8Q
  6. https://www.11x.ai/llms.txt
  7. https://docs.11x.ai/alice/setup
  8. https://www.11x.ai/products/alice/pricing
  9. https://docs.11x.ai/alice/knowledge-base
  10. https://docs.11x.ai/alice/playbooks
  11. https://docs.11x.ai/alice/icp-and-targeting
  12. https://docs.11x.ai/changelog
  13. https://www.11x.ai/launch-sequence/campaign-strategist-for-ai-outbound-success
  14. https://www.youtube.com/watch?v=Tn_WQ9mZbCI
  15. https://docs.11x.ai/alice/deep-research
  16. https://www.11x.ai/platform/data-lead-generation/lead-qualification
  17. https://docs.11x.ai/alice/ai-personalization
  18. https://docs.11x.ai/alice/sequences
  19. https://docs.11x.ai/alice/smart-replies
  20. https://docs.11x.ai/integrations
  21. https://docs.11x.ai/integrations/slack
  22. https://docs.11x.ai/data/company-and-lead-database
  23. https://www.11x.ai/compare/11x-vs-artisan
  24. https://docs.11x.ai/data/signals-and-triggers
  25. https://docs.11x.ai/data/website-visitor-tracking
  26. https://www.11x.ai/worker/alice
  27. https://www.youtube.com/watch?v=fegwPmaAPQk
  28. https://zenml.io/llmops-database/rebuilding-an-ai-sdr-agent-with-multi-agent-architecture-for-enterprise-sales-automation
  29. https://docs.11x.ai/security/audit-and-oversight
  30. https://docs.11x.ai/help-center/overview/what-results-to-expect
  31. https://docs.11x.ai/help-center/faq/billing
  32. https://www.11x.ai/legal/terms
  33. https://syncgtm.com/blog/11x-ai-review
  34. https://techcrunch.com/2025/03/24/a16z-and-benchmark-backed-11x-has-been-claiming-customers-it-doesnt-have/
  35. https://docs.11x.ai/integrations/webhooks
  36. https://www.11x.ai/platform/integrations/api
  37. https://docs.11x.ai/llms.txt
  38. https://deepline.com/gtm-stack/providers/11x
  39. https://techcrunch.com/2025/05/05/11x-ceo-hasan-sukkar-steps-down/
  40. https://www.trustpilot.com/review/11x.ai
  41. https://www.trustpilot.com/review/11x.ai?stars=1&stars=2
  42. https://www.youtube.com/watch?v=S7Sjpc6vzVY
  43. https://www.youtube.com/watch?v=da6Duod_fZo
  44. https://www.youtube.com/watch?v=wSaJA5jzHeY
  45. https://www.youtube.com/watch?v=Wxpvetyt8Do
  46. https://www.11x.ai/platform
  47. https://www.youtube.com/watch?v=qetWNHND3nk
  48. https://www.youtube.com/watch?v=bmOnFuYs7rA
  49. https://www.youtube.com/watch?v=FQwn-6eNopg
  50. https://www.11x.ai/launch-sequence

Kite / appsmith-v2 (7)

  1. https://github.com/appsmithorg/appsmith-v2/blob/main/backend/app/llm/skills/prospect-research/SKILL.md
  2. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/datasource-atlas/cards/contacts-router.md
  3. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/datasource-atlas/ROSTER.md
  4. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/datasource-atlas/CAPABILITY-MATRIX.md
  5. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/datasource-atlas/cards/clay-mcp.md
  6. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/datasource-atlas/README.md
  7. https://github.com/appsmithorg/appsmith-v2/blob/main/agent-context/research-enrichment-program/design-questions.md