Frontier system · Operations · 6 stages · updated 2026-10-23
Company brain: docs, data and tickets in Slack
A Slack agent that answers from docs and past threads with sources, answers data questions with SQL, files tickets from conversations, and reports gaps in your docs.
1Is this my problem?
The same questions get asked in Slack every week and answers are spread across Notion, Drive and old threads.
You'll recognise this if…
- Senior people answer the same internal question repeatedly.
- New hires take weeks to find things.
- Simple data questions wait days for an analyst.
2Is this solution worth it?
Your ROI, with the assumptions shown
Prefilled with our illustrative assumptions for a typical team (20 h/week of manual work at $70/h, 30% automated). Change them to your numbers.
Limits and when not to build it
Not worth it if your docs are thin or out of date; fix the docs first.
Guardrails it needs
- Respect document permissions.
- Read-only database role.
- Always cite sources.
3How does it work?
The agency builds you one internal tool, Company Brain, that runs the whole process. Watch it work on a sample case:
Pick a sample on the left and press Run example to watch Company Brain work through a case step by step.
Your tool connects to
Most of it is ordinary automation. AI agents handle 3 of 6 steps, and people approve at no checkpoints.
Under the hood: stages, vendors and the learning loop
AI agent Automation Human checkpoint
- 1Ingest
Index Notion, Drive, help centre and chosen channels with permissions.
Vector DBAutomation - 2Retrieve
Contextual retrieval and BM25 for fewer misses.
EmbeddingsAutomation - 3Answer
Answer with links; say "don't know" when unsure.
AI SDKClaudeGPTAI agentDecides: Answer or say "don't know" - 4Query data
Text-to-SQL against a read-only warehouse role.
BigQuerySnowflakeAI agentDecides: Write the SQL - 5Act
Turn a thread into a Jira or Linear ticket on request.
JiraLinearAI agent - 6Gap report
Weekly list of unanswered questions to doc owners.
SlackAutomation
Thumbs-down answers and unanswered questions become doc tasks and eval cases, so the bot is measured on real misses.
Example run
Illustrative sample data showing what moves between stages. Not taken from the sources.
- 1Ingest · in
Notion + Driveout12,400 chunks indexed - 2Retrieve · in
"What is our refund policy for annual plans?"out3 passages - 3Answer · in
PassagesoutAnswer with 2 links - 4Query data · in
"How many trials converted last month?"outSELECT ... -> 212 (read-only) - 5Act · in
ThreadoutLinear ticket ENG-881 created - 6Gap report · in
Weekout7 unanswered questions sent to owners
Vendors commonly used
Where AI decides and where people do
- Answer: Answer or say "don't know"
- Query data: Write the SQL
- Act: Turn a thread into a Jira or Linear ticket on request.
Maturity ladder
What each level adds. Most teams start at level 1 or 2 and climb. Highlighted: where this build sits.
- Level 1Basic automation
A wiki and a search box.
- Level 2AI-assisted
An LLM answers questions from pasted docs.
- Level 3Autonomous agent
A Slack agent retrieves with permissions, cites sources, runs read-only SQL and files tickets.
- Level 4Self-learning system
Thumbs-down and unanswered questions become doc tasks and eval cases.
Building blocks
Start with one of these smaller workflows and add stages.
A Slack bot answers team questions (lead stages, processes, policies) from internal documentation, with links, so people stop interrupting the one person who knows..
Who can build this →Team members ask questions in plain English; an agent writes SQL against your warehouse, runs it, checks the result and explains it, in several reasoning phases..
Who can build this →Seen in the wild
- vercel-labs/ai-sdk-slackbot (GitHub)
- Vercel: what we learned building agents
- Anthropic: Contextual Retrieval
- Zapier: how Zapier uses AI internally
Links checked 2026-10-23. The pipeline combines patterns from these public builds; we are not affiliated with these companies.
4Who can build it?
Matched on the services, platforms and industries agencies list on their own sites, and on their published projects.
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Related project: AI Automation Case Study: Edinburgh Boiler Company — Discover how flowio used AI Voice and SMS agents to cut lead response times from 1 day to 5 minutes for The Edinburgh Boiler Company.
Published price: From £750
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Related project: AI outbound sales platform case study — How SBD built IntellReach: an AI outbound sales platform with personalised emails, sequences, reply handling and human handoff, on Next.js, …
No published price
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Related project: Sample 3PL (illustrative): Invoice processing time cut 62% with an extraction agent — A 120-person logistics company replaced manual invoice keying with an extraction agent that reads PDFs and emails, matches them to purchase …
Published price: From $11,500
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Example project: AI & Software Development Case Studies — Explore THE TISA’s AI, SaaS, web, mobile, and custom software projects. See each project’s challenge, solution, and delivered product.
No published price
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Related project: Designing Two GPT-3 Products For O.C. Tanner — The popular HR company O.C. Tanner, which has been in business since 1927 and has over 1500 employees, was looking to research and design tw…
No published price
Why matched: RAG/knowledge base; Internal tools/admin panels; uses Claude.
Related project: AI Service Desk Software: A Copilot Build — AI service desk software that detects urgency, resolves L1 and L2 tickets, and works inside ServiceNow and Ivanti rather than beside them.
No published price
Projects, tools and prices come from each agency's own website (crawled October 2026). Not paid placements.
5What should I do next?
Questions
What is the company brain: docs, data and tickets in slack?
A Slack agent that answers from docs and past threads with sources, answers data questions with SQL, files tickets from conversations, and reports gaps in your docs.
Who is this for?
Companies of 30-500 people with knowledge in several tools and a busy Slack, often with a data warehouse.
When should we not build it?
Not worth it if your docs are thin or out of date; fix the docs first.
How does it improve over time?
Thumbs-down answers and unanswered questions become doc tasks and eval cases, so the bot is measured on real misses.
What keeps it safe?
Respect document permissions. Read-only database role. Always cite sources.
Who runs something like this?
Seen in the wild: vercel-labs/ai-sdk-slackbot (GitHub); Vercel: what we learned building agents; Anthropic: Contextual Retrieval; Zapier: how Zapier uses AI internally.
What does it cost to build?
Our rough estimate for a first production version with an agency: $10k–$40k over 6-12 weeks, plus tool and model usage.