Joined November 2023
- Tweets 318
- Following 103
- Followers 31
- Likes 68
111 Photos and videos
Pinned Tweet
24 Apr 2025
Recce 1.0 is now live on Product Hunt!
producthunt.com/posts/recce-…
Upvote and leave a comment to help us grow the Recce community and bring better data review processed to more data teams
Thanks for your support!
#OpenSource #Data #DataEngineering #Analytics #DeveloperTools #dbt
129
"A trillion dollars is arrayed against us." Stonebraker at 82, betting SQL should orchestrate data queries, not LLMs.
youtube.com/shorts/3PPwQViNy…
#AgenticAI #TextToSQL #SQL
8
NoSQL to "not only SQL" to "not yet SQL." Stonebraker's three-word history of a trillion-dollar industry detour. youtube.com/shorts/eATNtE37T…
#NoSQL #SQL #DatabaseHistory
4
"Don't ever bet against the compiler." Stonebraker on 40 years of database history and why NoSQL converged back to SQL. youtube.com/shorts/eduL-osMx…
#SQL #NoSQL #Postgres
13
"We tried out the LLMs that everyone else was touting. Accuracy was about 10%, not 80%." Stonebraker on real enterprise text-to-SQL. youtube.com/shorts/iMKrev7ZK…
#TextToSQL #LLM #DataWarehouse
13
Mike Stonebraker created Postgres, won the Turing Award, and just tested text-to-SQL on real warehouses. The result: 10%, not the 80% benchmarks claim. New episode.
heavybit.com/library/podcast…
#TextToSQL #Postgres #DataRenegades
7
Day 1 of #SnowflakeSummit is a lot. Booth visits, back to back talks, so many handshakes. Tomorrow night, relax with us. 🍻
Data Renegades Happy Hour, Tuesday June 2, 6-9pm, 7 min walk from the summit.
Drinks, not slides. 😉
luma.com/vnyf1nij?utm_source…
37
Data review flagged 99.999% row-count variance. PR was two lines.
Base: 5 years of prod history. Current: 1-hour CI build.
False alarms train reviewers to scroll past. That's the damage.
blog.reccehq.com/session-bas…
#dbt #DataEngineering
13
90% of enterprise programmers spend their time on maintenance, not greenfield development.
Michael Stonebraker's take on where AI actually earns its keep inverts the marketing narrative completely.
1
10
And listen at: heavybit.com/library/podcast…
1
6
Michael Stonebraker was right about CODASYL. Right about NoSQL. Now he's run text-to-SQL on a real enterprise warehouse and got 10% accuracy against an 80% benchmark.
The pattern is hard to ignore.
blog.reccehq.com/benchmarks-…
#DataEngineering #TextToSQL #AI
88
Before you let agents touch your codebase, build these gates.
Not because you don't trust the agent, but because you wouldn't trust anyone without them. Including yourself.
blog.reccehq.com/before-you-… #ClaudeCode #AIAgents #DevWorkflow
8
A wiki is something you look at. A shared AI system is something you work through.
When knowledge lives inside the workflow, it stays current. Every time someone runs a skill, outdated entries get noticed. Gaps get filled.
blog.reccehq.com/we-didnt-se…
#ClaudeCode #DataEngineering
30
AI coding tools generate plausible but wrong SQL constantly. The fix isn't waiting for a smarter model.
AI skills are markdown files that encode domain knowledge into coding tools. No framework, just structured text in a repo.
2
14
The loop: code → review → handoff → skills update. Every session makes the next one smarter. One aggregation bug became a permanent rule enforced automatically.
@data_dori broke it all down at Data Debug SF. Full writeup:
blog.reccehq.com/ai-skills-f…
#DataEngineering #AI
11
Our own Kent Chen wrote up the multi-agent architecture the team built for Recce's AI Data Review.
Single agent kept forgetting findings as PRs got complex. Fix: orchestrator two specialists, each with its own 200k context window.
1
14
One subagent fetches full PR context via a single GitHub GraphQL MCP call (replaced 5-10 gh CLI round-trips). The other explores data through 6 Recce MCP tools: lineage_diff, schema_diff, row_count_diff, custom queries.
1
12
Subagents return summaries to orchestrator, not raw payloads. Built with Claude Agent SDK and MCP.
Read full post here: blog.reccehq.com/designing-r…
#dbt #DataEngineering #AI #MCP
37