For the head of product · every figure at 20 seats · free, no signup

5 AI tools at 20 people: $47,520 a year

That is the median row in this index — $9,504/year a tool at 20 seats — multiplied by 5, or $2,376 a year per person. It is not a survey and not an estimate: it is arithmetic on the 65 researched prices in the table below, each one read off the vendor’s own pricing page at the seat count a product org this size actually lands on.

What your stack costs a year, at 20 people

AI tools you pay forCheaper quarter$4,800 eachMedian tool$9,504 eachDearer quarter$24,000 each
3 tools$14,400$28,512$72,000
5 tools$24,000$47,520$120,000
8 tools$38,400$76,032$192,000

The three columns are the lower quartile, the median and the upper quartile of the 65 researched prices below, all at 20 seats. Which 5 you picked is the whole spread: the 5 dearest rows here come to $209,000/year together (Decagon Duet Apprentice, Conveo, Ranger, Kubit, Intercom Fin 3), the 5 cheapest to $7,688/year. And 6 of these vendors publish no rate card at all, so part of any real bill is a number you can only get by asking.

Price your own stack instead → — tick the tools you actually pay for, set your own headcount, and get the annual total with every figure’s read date and a link to the vendor page it came from.

$47,520 of tooling. $96/year to know what to cut.

Signal Brief is five new AI tools every weekday morning, each with the price at 20 people, an adopt / pilot / skip verdict, the SSO, SOC 2 and data-handling line, and three lines you can paste into #product. $10/month with 7 free days first, $96/year paid once, or $490/year for the Desk Licence — unlimited readers · one organisation · one invoice.

Seen enough · no email form · card checkout

Unlock all 108 verdicts now.

The locked column on every one of these 108 rows, plus five new tools with the brief on each, every weekday at 7am. No email form — Stripe takes the address your issues go to at the card step.

  • A one-word verdict on every tool — adopt, pilot or skip — with the reason in one sentence.
  • The real price a 20-person product org lands on, not the headline tier on the pricing page.
  • The integration and data-risk line: which plan SSO needs, the SOC 2 posture, where your roadmap and customer data ends up.
  • A paste-ready 3-line pitch for #product, so making the case costs you nothing.
$96/year
paid once · 12 months of briefs · or $10/month

Stripe takes the address your issues go to at the card step · $96/year is one charge for 12 months · $10/month starts with 7 free days · cancel any time by emailing [email protected]

For a head of product buying for the whole org
$490/year
unlimited readers · one organisation · one invoice · paid once for 12 months

Your PMs, your designer and your EM all read the same brief before standup. A Desk Licence covers every reader at one organisation on one invoice — no seat count to pick, no seat admin, nobody to remove when a PM leaves.

Against the $96/year plan it breaks even at five readers: $490/year for the licence against $480 for five annual subscriptions. From the sixth reader on it is the cheaper of the two, and the tenth and fortieth cost nothing more.

We re-read 46 vendor pricing pages every morning — 45 tools, 71 published figures — and diff them against the figures we published.

Most recent change: 2026-10-23 — Fin.

See every dated change, with the vendor's own excerpt →

One charge, one receipt · no seat admin · reply to the first issue with the addresses to add and they are on tomorrow's send · questions first? [email protected]

7 weekday issues free · no card · nothing renews

Or read 7 issues free first.

The five tools are published free every weekday morning. Subscribers also get the Adoption Brief on each one: adopt / pilot / skip, the price a 20-person product org really lands on, the SSO and data-risk line, and a pitch you can paste into #product. Leave your email and the next 7 issues arrive with all five briefs attached — no card, nothing to cancel.

  • A one-word verdict on every tool — adopt, pilot or skip — with the reason in one sentence.
  • The real price a 20-person product org lands on, not the headline tier on the pricing page.
  • The integration and data-risk line: which plan SSO needs, the SOC 2 posture, where your roadmap and customer data ends up.
  • A paste-ready 3-line pitch for #product, so making the case costs you nothing.

Read a real Adoption Brief →

Free for 7 issues
no card · no password · nothing renews

Rather skip the trial and subscribe now? $10/month, or $96/year paid once.

No card to start the trial · 7 weekday issues with every brief · then the five tools stay free · $10/month or $96/year keeps the briefs · stop any time by emailing [email protected]

Rather look first? Every figure above is worked out in the open: all 108 rows, dearest first, one complete brief, unredacted, and the whole table as CSV — free, no email, no sponsors.

All 108 rows, dearest first

Every tool below is one we have covered and priced at a 20-person product org — the number the vendor hides behind “Contact sales”. Read off their own pricing page at the seat count a real team buys where they publish one, and marked buyer-reported or our own arithmetic where they do not. Not the headline tier, and not a list price nobody pays. The spread runs from $1,188/year to $50,000/year.

1 of the 65 researched prices below is a figure no one else publishes. Checked 2026-10-21: for that row there is no vendor rate card and no third-party page carrying a price for the tool at all, so this table is the only published source for it. Flagged in the table below. How that was checked →

No sponsorships · no affiliate links · no vendor has ever paid to be in this table

65 researched prices — 2 published by the vendor, 21 buyer-reported, 42 our own arithmetic on published units — plus 6 vendors who publish nothing and 37 tools we have not priced yet: 108 rows in all. Every figure is what a 20-person product org lands on for the issue it ran in, with the basis printed under it, never a headline tier and never an estimate we did not show the working for.
ToolCategoryWhat 20 people payVerdict
Decagon Duet ApprenticeDuet Apprentice learns from internal docs and escalated conversations, then drafts and updates agent operating procedures from Slack or Teams threads.Support agent ops$50,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
ConveoRuns asynchronous AI-moderated video interviews with adaptive follow-up questions in 50+ languages, then themes and clips the transcripts.AI-led customer interviews$45,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
KubitKubit runs product analytics and agent-observability queries directly in Snowflake, BigQuery or Databricks without copying event data out.Warehouse-native analytics$40,000–$100,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
RangerRanger's agents write, run and maintain Playwright tests against your web app, and now re-test fixes produced by coding agents.Autonomous QA testing$40,000–$90,000/yearbuyer-reportedOnly published sourceAdopt / Pilot / Skip — lockedLocked
Intercom Fin 3Customer service agent with Procedures for multi-step policies, Simulations for pre-release testing, and deployment across email, chat, voice, Slack and Discord.Support resolution agent$34,000–$40,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Amplitude AI AgentsA set of agents plus a Global Agent that query product data in natural language, run analyses, and surface changes without a dashboard.Product analytics agents$30,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Bug0Bug0 pairs an open-source agentic Playwright engine with a forward-deployed QA engineer who builds tests, triages failures, and gates releases in your CI.Managed AI QA$30,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Enterpret Agent OSAgent OS runs Enterpret's feedback agent continuously, classifying incoming feedback and triggering automations without someone starting the analysis.Customer feedback intelligence$30,000–$100,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
FinIntercom's third-generation support agent answers customer conversations across channels, runs configured procedures and hands off to humans when it cannot resolve.vendor price moved 2026-10-23AI support agent$30,000–$45,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Sprig Analyze AgentSprig's Analyze Agent, in preview since October 8, reads a study's survey and session data and returns findings with linked evidence.Product research analysis$30,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
GoodData.AI Agentic Serving PlaneA governed execution layer that serves high-concurrency, context-rich metric queries to AI agents with access controls and contextual grounding, announced October 6.Governed analytics serving$29,940/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
New Relic AI EvaluationScores model and agent outputs in production and links quality drops to traces, model versions and token cost.AI feature observability$28,000–$40,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
LorikeetMulti-agent support system that runs multi-step workflows inside Zendesk, Intercom or Salesforce, resolving tickets end-to-end and escalating when uncertain.Support resolution agent$25,200/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Fullstory StoryAI AgentsAgents inside Fullstory that investigate reported issues, analyze funnel drop-off, assess release impact and monitor sessions continuously, reporting findings to Slack or email.Behavioral analytics agents$25,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
LaunchDarkly AgentControlManages prompts, models and agent configs at runtime, with offline and online evals, guarded rollouts and LLM tracing.AI feature release$25,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Listen LabsRuns AI-moderated voice and text interviews with recruited or self-sourced participants and turns the transcripts into segmented findings.AI user research$25,000–$40,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Confident AI Customer MonitoringCustomer Monitoring, launched in Confident AI's Q3 launch week, breaks LLM traces and eval scores down by account so you can see per-customer AI quality.AI feature analytics$24,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Inkeep AgentsOpen-source framework and visual builder for multi-agent assistants that answer from docs and help content, with managed retrieval on Enterprise.Docs and support agents$20,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Pendo NovusInstruments a web app from one CLI command, then posts goal-tied findings, session replays and drafted fixes into Slack each morning.Product analytics agent$20,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
SpurAI agents that run end-to-end web and native mobile tests written in plain English, on pull requests or on a schedule.Agentic QA testing$20,000–$60,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
StrellaRuns AI-moderated voice interviews with your customers or an 8M-person panel, then synthesizes themes, transcripts and highlight reels automatically.AI customer research$20,000–$40,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
PylonInvestigates each incoming ticket across connected systems before a human opens it and can take approved actions or delegate to other agents.B2B support triage$16,700/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Front CompassCompass scores every conversation handled by Front Autopilot or externally built agents on tone and comprehension, with inferred CSAT and improvement recommendations.Agent quality scoring$15,600/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Maze AI ModeratorBuilds a discussion guide from a stated research goal, then runs moderated interviews with participants automatically and themes the transcripts.Automated user interviews$15,000–$30,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Zendesk AI Agent Conversation SimulatorZendesk's simulator, released this month, generates realistic customer conversations from scenarios you write and runs your AI agent against them before deployment.Support agent testing$13,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
QA Wolf Mapping AIExplores a web app autonomously and outputs a structured, plain-English list of workflows and test cases, updating it as features ship.Test coverage mapping$13,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Dovetail Digital TwinsBuilds queryable models of customer segments from your own calls, tickets, surveys and research, with each answer linked back to its source.Customer discovery$12,000–$30,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
PostHog AIAn analytics agent in PostHog that writes SQL, builds multi-step analyses, and creates dashboards, surveys, alerts, and feature flags from a plain-language question.Product analytics agent$12,000–$20,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Productboard SparkAn AI agent inside Productboard that reads feedback, drafts discovery documents, links insights to features, and answers questions about your product data.Feedback and roadmap$12,000–$24,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Mixpanel AIMixpanel AI adds an in-product agent that answers analytics questions, diagnoses metric movements and runs report tasks across your event data.Product analytics agent$10,080–$19,620/yearvendor-publishedAdopt / Pilot / Skip — lockedLocked
Graphite AgentReviews pull requests and answers questions about a diff inside GitHub, replacing Graphite's earlier standalone Diamond reviewer.AI code review$9,600/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
KiroKiro is AWS's agentic IDE and web agent that turns a written spec into requirements, design and tracked implementation tasks.Spec-driven development$9,600/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Plain AriAri is Plain's customer-facing AI agent that answers and resolves B2B support threads in Slack, email, Teams and in-app chat.Support triage agent$9,504/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Hex ThreadsA chat interface over your warehouse that answers data questions using semantic models curated by the data team, shipped in Hex's Fall 2025 launch.Conversational product analytics$9,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
PostHog AIChat-based agent inside PostHog that builds insights and dashboards, writes SQL, summarizes session recordings and answers product data questions.Product analytics agent$9,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Unified.to Agent SkillsPublishes 100+ SKILL.md files that teach coding agents to set up vendor OAuth apps and finish integration work, tracked as typed tasks.Integration automation$9,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
LaunchDarkly AgentControlConfigures prompts and models behind flags, runs offline and online evals, and now drives experiments end to end through an MCP server and agent skills.AI release control$8,000–$12,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Pendo for AgentsCombines Agent Toolkit, which feeds live user context to in-app agents, with Agent Analytics, which captures prompts, use cases and failure signals.In-app agent analytics$8,000–$25,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Productboard SparkA product-management agent inside Productboard that works from your workspace feedback, roadmaps and documents to run discovery, competitive analysis and spec drafting.AI product agent$8,000–$20,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
Canny AutopilotReads support chats, sales calls and reviews, extracts feature requests, deduplicates them and routes each one to the right team.vendor figure changed 2026-10-20Feedback triage$7,900/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
MachaPuts an AI agent on chosen ticket categories inside Zendesk, Freshdesk, Gorgias, Front or Intercom, resolving threads end to end using your past tickets and APIs.Support ticket automation$7,200–$14,400/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
TestCollab AI AgentsA directory of scoped QA agents for regression, smoke, API, accessibility and performance checks that run on webhooks or schedules and return evidence.Autonomous QA agents$7,020/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Dovetail AI AgentsAgents in Dovetail's fall release monitor incoming feedback channels, run analysis on the data, and post findings to Slack, Teams, or Linear.Customer insight automation$7,000–$10,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
OpenAI AgentKitVisual Agent Builder, embeddable ChatKit interface, connector registry and evals for building and shipping agents on OpenAI models.Agent building platform$6,000–$24,000/yearbuyer-reportedAdopt / Pilot / Skip — lockedLocked
QA.techQA agents crawl a web or mobile app, generate tests from natural-language goals, and run them on pull requests.Agentic QA testing$6,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
cubiccubic reviews pull requests for bugs and context violations, can open fix PRs, and maintains an auto-updating wiki of the codebase.AI code review$5,400/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Hex Notebook AgentAgent inside Hex notebooks that writes SQL and Python, runs multi-step analyses, and leaves the generated code visible for review.Agentic product analytics$5,400/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Mintlify AssistantAn agent in the Mintlify dashboard that writes documentation from prompts or merged pull requests and updates existing pages, code samples and API references.Documentation writing agent$5,400/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
AlloyAlloy clones an existing web app, then lets you change it with prompts or visual edits and share the result as an interactive prototype.AI prototyping$4,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Figma AgentFigma's agent left beta on October 6 and can search files, follow uploaded library guidelines, and build or edit designs from a prompt.vendor price moved 2026-10-18Design agent$4,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Miro for Product AccelerationMiro's new product suite puts AI agents on the canvas to turn research, sticky notes and sessions into briefs, specs and prioritised backlogs.Product planning canvas$4,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Notion AgentsNotion 3.8 connects workspace data, external tools and skills to agents, and lets agents propose page changes instead of making them directly.Docs and context$4,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Vercel AgentAn agent in Vercel that reviews pull requests with sandbox-validated suggestions and investigates deployment failures, runtime errors and anomaly alerts.Code review and incidents$4,800/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Atlassian RovoRovo searches and chats across Jira, Confluence and connected tools like Figma, GitHub and Zendesk, and runs configurable agents on that context.Connected work search$4,780/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Linear Product IntelligenceA Linear capability that automates routine product-operations work such as triaging incoming issues, deduplicating requests and summarizing project state.Product operations automation$3,840/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
FrankRuns AI-moderated voice and chat customer interviews asynchronously, asks follow-up questions, then returns transcripts, themes and sentiment analysis.AI customer interviews$3,750/yearvendor-publishedAdopt / Pilot / Skip — lockedLocked
LangSmith Insights AgentReads production agent traces and groups full multi-turn conversations into labelled usage categories with latency, volume, and eval scores per category.AI feature analytics$3,700–$4,700/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
ChatPRDDrafts PRDs and specs from shared project context, with a Linear agent that fills issue descriptions and a hosted MCP endpoint for IDEs.Spec and PRD drafting$3,490/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
BraintrustAn agent inside Braintrust that reads production logs, writes test cases and scorers, and runs evaluations on your LLM features.AI feature evals$2,988/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Sentry SeerAI debugging agent that scans production errors, performs root-cause analysis using Sentry telemetry and opens pull requests with proposed fixes.Bug triage and debugging$2,880–$9,600/yearour arithmeticBilled per active contributor (2+ PRs a month to a Seer-enabled repo), not per seat: the floor assumes six of the twenty qualify, the ceiling assumes all twenty do.Adopt / Pilot / Skip — lockedLocked
Lovable CloudAdds a provisioned backend — auth, database, storage, AI calls — to Lovable prompt-built apps without wiring separate infrastructure.Prototyping with backend$2,000–$6,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
KojiRuns AI-moderated voice or text interviews with your users and compiles the transcripts into themed, quotable reports.AI customer interviews$1,500–$3,500/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
MomenticMo is an agent that explores your app from a written brief, runs flows in hosted browsers and devices, and files bugs with repro steps, logs and video.AI QA agent$1,500–$6,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Replit Agent 3Builds and tests running web apps from a prompt, working autonomously for extended sessions and fixing its own failures in the browser.Prototyping agent$1,500–$3,000/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Perspective AIRuns AI-moderated conversational interviews with your customers at scale and produces summaries, themes and shareable highlight clips.AI customer interviews$1,188/yearour arithmeticAdopt / Pilot / Skip — lockedLocked
Dataiku Agent ManagementDiscovers AI agents running across Bedrock, Databricks, Vertex, Copilot Studio and Agentforce, tracks their KPIs and ranks them by risk.AI agent governanceVendor publishes noneAdopt / Pilot / Skip — lockedLocked
FernFern generates a docs site and client SDKs from an OpenAPI spec and ships AI search, an AI writing agent, and an MCP server for the docs.API docs and SDKsVendor publishes noneAdopt / Pilot / Skip — lockedLocked
OutsetOutset runs hundreds of AI-moderated interviews at once by video, voice or text, screens out low-quality participants, and synthesizes the transcripts.AI-moderated researchVendor publishes noneAdopt / Pilot / Skip — lockedLocked
UviUvi builds a permissioned knowledge graph across Slack, Jira, Confluence, code, coding sessions and meeting notes so AI tools can trace past decisions.Decision context layerVendor publishes noneAdopt / Pilot / Skip — lockedLocked
Virtuoso TouchstoneTouchstone reads requirement documents, proposes tests a named person approves, runs them on a deterministic engine and records every decision as evidence.Agentic QA platformVendor publishes noneAdopt / Pilot / Skip — lockedLocked
WSO2 Agent ManagerOpen-source control plane that registers AI agents, gives each an identity, applies guardrails and collects OpenTelemetry traces across any framework or model.Agent control planeVendor publishes noneAdopt / Pilot / Skip — lockedLocked
Ada.imUploads spreadsheets or connects databases and generates cleaned analysis, charts, and a written report without SQL.AI data analysisNot priced yetAdopt / Pilot / Skip — lockedLocked
Agent Compass by Future AGIBuilds a trace graph of agent runs to locate where a multi-step AI workflow failed and why.Agent failure diagnosisNot priced yetAdopt / Pilot / Skip — lockedLocked
Agnost AIProduct analytics for chat and voice agents that reads production conversations and flags behavioural failures — users cursing at the agent, rephrasing the same request, correcting it, or leaving after a technically successful answer.Agent analyticsNot priced yetAdopt / Pilot / Skip — lockedLocked
AmbralAmbral unifies CRM, support, meeting, and warehouse data into per-account models that flag churn risk and expansion opportunities daily.Post-sale revenue signalsNot priced yetAdopt / Pilot / Skip — lockedLocked
Ari by ArisoConnects to Slack, Linear, GitHub and meetings to track commitments, decisions and risks, then reports what changed and what has slipped.Execution trackingNot priced yetAdopt / Pilot / Skip — lockedLocked
AtlaAnalyzes agent traces to detect step-level errors, cluster them into recurring failure patterns, and suggest fixes you can test in-product.AI agent evaluationNot priced yetAdopt / Pilot / Skip — lockedLocked
AUDR by ChargebeeOpen JSON spec and SDKs that record who initiated an agent run and what each step of it cost.Agent cost standardNot priced yetAdopt / Pilot / Skip — lockedLocked
Basedash AgentAnswers plain-English questions against your connected databases and warehouses, returning governed charts and daily surfaced patterns.AI data analystNot priced yetAdopt / Pilot / Skip — lockedLocked
CapalyzeCapalyze scrapes live web data — reviews, listings, comments — into a spreadsheet engine you can query in plain language and chart.Web data analysisNot priced yetAdopt / Pilot / Skip — lockedLocked
CortexOpen-source tool that turns OpenAPI, GraphQL, gRPC and AsyncAPI specs into interactive docs, typed SDKs in 11 languages, and MCP servers.API docs generationNot priced yetAdopt / Pilot / Skip — lockedLocked
Decagon Voice 2.0Launched today alongside AOP Copilot and Simulations, it runs phone support agents and tests changes against simulated customer conversations before release.Support agent opsNot priced yetAdopt / Pilot / Skip — lockedLocked
Documentation.AIAn agent reads your repo, support tickets, and feedback after merges, then drafts documentation updates for human approval.Self-updating product docsNot priced yetAdopt / Pilot / Skip — lockedLocked
Dovetail — Digital Twins and AgentsThe Summer ’26 release adds "digital twins" built from your real calls, tickets and research that you can interrogate as if they were a specific customer or segment, plus background agents that route findings unprompted.Customer researchNot priced yetAdopt / Pilot / Skip — lockedLocked
EmbedfulConnects to your existing database and generates secure multi-tenant dashboards you embed in your product with short-lived JWT tokens.Embedded customer analyticsNot priced yetAdopt / Pilot / Skip — lockedLocked
Gamma 3.0Adds Gamma Agent for prompt-driven editing of decks and docs, smart diagrams, and an API for generating documents programmatically.Stakeholder communicationNot priced yetAdopt / Pilot / Skip — lockedLocked
GhostshipGhostship runs AI agents against your app from a URL and a described user journey, reporting bugs as a GitHub CI check.Automated QA agentsNot priced yetAdopt / Pilot / Skip — lockedLocked
GitHub Spec KitOpen-source toolkit that turns a written specification into a structured plan, tasks, and implementation steps for AI coding agents.Spec-driven developmentNot priced yetAdopt / Pilot / Skip — lockedLocked
GrapevineGrapevine indexes Slack, Google Drive, Linear and your codebase, then answers company-specific questions in Slack with citations.Internal knowledge searchNot priced yetAdopt / Pilot / Skip — lockedLocked
GrowthBook 5.0 Product AnalyticsGenerally available on 22 July: funnels, dashboards, metric explorer and an AI chat agent, all running on the same metric definitions that already power your feature flags and experiments.Analytics and experimentationNot priced yetAdopt / Pilot / Skip — lockedLocked
InterfereInterfere monitors a live app for behavior changes, identifies the root cause in code, and tracks each issue to resolution.Production issue detectionNot priced yetAdopt / Pilot / Skip — lockedLocked
KraftfulAggregates feedback from support tickets, reviews, surveys and calls, then clusters it into recurring requests and complaints.Feedback analysisNot priced yetAdopt / Pilot / Skip — lockedLocked
Linear for AgentsLets you assign AI agents to Linear issues so they triage incoming reports, answer questions, and open follow-up work.Issue triage agentsNot priced yetAdopt / Pilot / Skip — lockedLocked
Magic Mic by GuiddeRecords you walking through a workflow while talking, then generates a narrated, step-by-step video guide with cleaned-up voiceover.Product documentation videoNot priced yetAdopt / Pilot / Skip — lockedLocked
Murmur LabTracks conversations across Reddit, Discord, X and niche forums, then produces reports on why a topic is gaining traction.Social listening researchNot priced yetAdopt / Pilot / Skip — lockedLocked
PieRuns AI agents that explore your web, iOS and Android app, generate and maintain tests, and report functional and usability bugs.Autonomous QA agentsNot priced yetAdopt / Pilot / Skip — lockedLocked
PleomPleom connects warehouses, CRMs, and support tools, answers plain-English data questions, and pushes anomaly alerts with suggested actions.Agentic business intelligenceNot priced yetAdopt / Pilot / Skip — lockedLocked
ProductBridgeCombines live chat, help center, surveys and a feedback board so support conversations become deduplicated, reach-and-revenue-scored feature requests.Support and feedbackNot priced yetAdopt / Pilot / Skip — lockedLocked
Replay QAPoint it at a staging or localhost URL and it derives user journeys, drives the app, and reports bugs from deterministic recordings.Autonomous QA testingNot priced yetAdopt / Pilot / Skip — lockedLocked
Sentry AI Code ReviewAnnounced today in beta, it reviews GitHub pull requests using Sentry's runtime error and performance data, predicts breakages, and generates tests.Pre-production qualityNot priced yetAdopt / Pilot / Skip — lockedLocked
Snap by VersiveRuns usability tests on Figma prototypes, live sites, or screenshots using AI personas instead of recruited participants, returning results in minutes.Rapid usability testingNot priced yetAdopt / Pilot / Skip — lockedLocked
Specter — Competitor AgentA one-click agent that maps any company's competitive landscape in under a minute, separating direct from adjacent competitors, each with funding, valuation and revenue estimates.Competitive intelligenceNot priced yetAdopt / Pilot / Skip — lockedLocked
Sprinto AI Security QuestionnaireDrafts answers to inbound security questionnaires and RFPs using your existing compliance evidence and past responses.Security review automationNot priced yetAdopt / Pilot / Skip — lockedLocked
Strata by Klavis AIStrata is a single MCP server that lets AI agents discover and call tools across many apps step by step instead of loading every tool at once.Agent tool infrastructureNot priced yetAdopt / Pilot / Skip — lockedLocked
TheneoTheneo generates, lints and publishes API reference docs from your spec, and keeps them updated as endpoints change.API documentationNot priced yetAdopt / Pilot / Skip — lockedLocked
UxiaUxia runs synthetic AI testers through your user flows or screens and returns friction points and suggested fixes in minutes.AI usability testingNot priced yetAdopt / Pilot / Skip — lockedLocked
VotitoLaunched 20 July out of London, it pairs public or internal feedback boards with structured survey methods, scoring ideas on votes, discussion and collected evidence rather than raw volume.Feedback and prioritisationNot priced yetAdopt / Pilot / Skip — lockedLocked
Zeplin AI Design ReviewReviews screens in Zeplin for missing states, inconsistent components and gaps that would block developer handoff.Design handoff reviewNot priced yetAdopt / Pilot / Skip — lockedLocked

“Only published source” means we could find no price for that tool published anywhere but here — not on the vendor's own rate card, and not on any third-party page, tracker, round-up or directory listing. How it was decided, on 2026-10-21: we took every researched row the price watch can find no vendor-published figure for and that no page in our citation ledger cites (23 rows), searched each one by hand — the tool's name with "pricing" and "cost", the figure itself, and its G2, Capterra and Vendr listings — and read every result back. 22 of the 23 turned out to have a price published somewhere else, so they are not flagged, whatever our own records said.

The locked column is the Adoption Brief half: the adopt / pilot / skip call, the SSO, SOC 2 and data-handling line, and the three-line pitch you can paste into #product. It is what a subscription buys, and it arrives on five new tools every weekday morning — 7 issues of it with no card, or $96/year outright with no email form at all, both at the top of this page. Prefer one tool at a time? Each tool’s own page carries its write-up and the working behind the figure.

One row, unlocked, so you can see exactly what is behind the column

Here is The Adoption Brief for one tool in the table above, in full and unredacted: the verdict, the one sentence behind it, the price, the SSO, SOC 2 and data-handling line, and the three lines you can paste into #product. It is from issue #028, which is already published in full — nothing here is being given away that you could not already read.

01 · Product planning canvas · free to read

Miro for Product Acceleration

Miro's new product suite puts AI agents on the canvas to turn research, sticky notes and sessions into briefs, specs and prioritised backlogs.

Why it matters for PMsIt changes the week after a discovery workshop: synthesis and the first draft of the brief come out of the board instead of your calendar.

miro.com→

The Adoption Brief · normally subscribers only
Pilot

Cheap enough to test at $4,800 a year, but the AI value depends entirely on whether your discovery actually happens on a Miro board.

Real price for a 20-person product org
Business plan, not Starter: SAML SSO starts at Business per Miro's plan comparison. 20 members at $20/member/month billed annually = $4,800/year, plus the Miro AI add-on if credit limits bite. SCIM provisioning and data-governance controls require Enterprise, which is quote-only with a roughly 30-seat minimum.
Integration & data risk
SAML SSO on Business, SCIM only on Enterprise; SOC 2 Type II and ISO 27001 published on Miro's trust page; boards and AI prompts sit in Miro's cloud with LLM sub-processors, and Miro states customer content is not used to train models.
Paste this into #product
Proposing we trial Miro for Product Acceleration as the home for discovery workshops, so synthesis and the first brief draft come off the board itself.

Today a workshop costs us two to three days of manual sticky sorting and writing afterwards; this collapses that into a reviewable draft the same day.

Ask: one squad, two discovery sessions, one PM owning the board template, four weeks, then we compare brief quality against our last two.
Read off Miro for Product Acceleration's own pricing and trust pages when issue #028 was written, Monday, 26 October 2026. Prices move — check the vendor's page before you quote one internally.

That column is locked on the other 107 rows in this table. Unlock it — $96/year, $10/month with 7 free days, or $490/year for the whole organisation.

How these numbers were worked out

Quoting this table, or recomputing it

Every row above is also published as data, free and without an email. The whole table is one CSV at https://signal-brief19.acoco.ai/ai-tool-prices.csv — 18 columns: tool, category, one_line_description, price_status, price_label, amount_usd_low, amount_usd_high, price_period, annual_usd_low, annual_usd_high, annual_basis, seats, vendor_url, signal_brief_url, issue_number, issue_url, issue_date, price_checked_on. The same rows are embedded in this page as schema.org JSON-LD: a Dataset whose distribution is that CSV, and an ItemListcarrying each tool’s name, category and annual figure.

The table also resolves as one citable dataset. Its Croissant descriptor (/ai-tool-prices/croissant.json, ML Commons 1.1 — the file Google Dataset Search and Hugging Face read) defines every column and its type against the CSV above. The same schema is published as a Frictionless Data Package (/ai-tool-prices/datapackage.json), and a catalog document (/ai-tool-prices/index.json) carries the revision date, the terms and every distribution URL in one place. All three are regenerated with the table, so the version they state is the day the figures last changed — never a build date.

Read annual_basis before you quote a number — vendor_annual_figure = the vendor's own published yearly price; buyer_reported_annual_figure = a figure buyers, resellers or a third-party tracker report, which the vendor does not publish; signal_brief_arithmetic = our own multiplication on the vendor's published units; annualised_from_monthly_by_signal_brief = the vendor's published monthly price × 12; period_unstated = the brief quotes a figure with no period and we will not call it annual; none = no figure. An empty price is never a zero price: price_status says whether the vendor publishes nothing or we have not priced it yet. Each price is read off the vendor's own published pricing page, where there is one, on the day that tool ran in Signal Brief — or, for a row re-read by hand since, on the day of that re-read, which is the day price_checked_on carries — priced at 20 seats — the tier a 20-person product org actually lands on, including minimums and platform fees — never the headline tier and never a list price nobody pays. Where a vendor publishes no rate card, the row says so instead of carrying an estimate. Where a vendor publishes only a monthly figure, the annual column is that figure × 12 and annual_basis says so. Where the figure is a band buyers or third-party trackers report rather than one the vendor publishes, or our own multiplication on published units, annual_basis says that too — and a figure quoted for one seat, one test or one ticket is never published as the org's annual cost. Every published figure is checked before it is printed: its twelve-month equivalent must fall between $50 and $1,000,000 a year at a 20-person product org, and both ends of it must appear in the stored brief the row links to. A figure that fails either check is not published at all — the row says the vendor publishes no rate card, or that we have not priced the tool yet. Prices rot, so never present a figure without the issue date on its row.

Licence: CC BY 4.0. This dataset is published under the Creative Commons Attribution 4.0 International licence (CC-BY-4.0) — free to quote, redistribute and recompute, commercially or otherwise. The one condition is attribution: credit Signal Brief and link https://signal-brief19.acoco.ai/ai-tool-prices. Cite it as — Signal Brief, “AI tool price index” (https://signal-brief19.acoco.ai/ai-tool-prices), revision 2026-10-26, CC-BY-4.0. Found a figure that has moved, or want the row corrected? Email [email protected].

65 of these came out of 28 issues, five tools at a time, at 7am on a weekday. Read every issue in full, or browse the tools A–Z. The dearest line in the table is $50,000/year (Decagon Duet Apprentice, decagon.ai).

Signal Brief · the only thing we sell

Tomorrow morning, five more — with the call on each.

Every issue is published here in full, free, and stays that way. What a subscription adds is the Adoption Brief on all five tools: adopt, pilot or skip and why, the price a 20-person product org really lands on, the SSO and data-risk line, and a 3-line pitch you can paste into #product.

7 weekday issues free · no card · nothing renews
  • A one-word verdict on every tool — adopt, pilot or skip — with the reason in one sentence.
  • The real price a 20-person product org lands on, not the headline tier on the pricing page.
  • The integration and data-risk line: which plan SSO needs, the SOC 2 posture, where your roadmap and customer data ends up.
  • A paste-ready 3-line pitch for #product, so making the case costs you nothing.

Read a real Adoption Brief →

Free for 7 issues
no card · no password · nothing renews

Rather skip the trial and subscribe now? $10/month, or $96/year paid once.

No card to start the trial · 7 weekday issues with every brief · then the five tools stay free · $10/month or $96/year keeps the briefs · stop any time by emailing [email protected]

$10/month or $96/year after 7 free days, cancel in one click · a real Adoption Brief, in full · every issue, published free · [email protected]