Skip to main content

​
1. New dev onboarding

Before

3 days of “ask Sarah, she knows.” New devs piece together tribal knowledge from stale wikis and outdated READMEs. Sarah is in meetings all day.

After

Open the project board. Ask AI “how does auth work here?” Get an answer citing the actual architecture doc your team uploaded last month.
New dev asks: "How does auth work in this project?"

AI responds:
  "Based on the architecture doc uploaded by Sarah (March 2):
   We use Supabase Auth with JWTs. The middleware validates tokens
   in src/middleware/auth.ts. RLS policies handle row-level access.
   See the full doc on the board."
The new dev is productive on day 1, not day 4.

​
2. Shared team knowledge

Before

Someone finds a critical blog post about a production issue pattern. Shares it in Slack. It’s buried under 200 messages in 2 days. Nobody can find it again.

After

Add it to the board. The content is vectorized. Every dev and every AI assistant on the project can find it forever.
Dev: "Search the board for connection pooling issues"

AI responds:
  "Found 2 relevant sources:
   1. Blog post 'PgBouncer gotchas in serverless' (added by Jake, March 5)
   2. Your team's incident postmortem from Feb 28
   Both mention setting pool_mode=transaction. Here's the summary..."
Knowledge survives Slack. It survives team changes. It’s always findable.

​
3. Learning a new technology

Before

Watch a 45-minute YouTube tutorial. Forget 80% within a week. Search for the same thing again. Rewatch parts of it. Repeat.

After

Add the YouTube video to the board. The transcript is vectorized automatically. Ask AI about any part of it, anytime.
Dev: "What did that Inngest tutorial say about retry configuration?"

AI responds:
  "From the video 'Inngest Deep Dive' (timestamp 23:14):
   Set maxRetries in the function config. Default is 3.
   Use backoff: 'exponential' for API calls.
   The presenter recommended starting with 5 retries for webhooks."
Watch once. Remember forever.

​
4. Working in Claude Code

Before

Claude Code has no idea about your project’s tribal knowledge. It gives generic answers. You paste context manually every time.

After

Claude Code searches your team’s ChatGrid board. Answers are grounded in your actual docs, tutorials, and decisions. No copy-pasting.
You (in Claude Code): "How should I handle rate limiting for our API?"

Claude Code searches your ChatGrid board automatically.

Claude Code responds:
  "Based on your team's architecture decisions board:
   You're using Inngest's built-in rateLimit (set in planAgent.ts).
   The current limits are 100/day/user for plan agents and
   50/day/agent for email agents. The decision doc from March 4
   explains why these numbers were chosen."
Your AI assistant knows what your team knows.
To set up MCP in Claude Code, see the MCP Setup guide.