Recce - Trust, Verify, Ship

"If you have a successful career built on glazing people, I kind of hate you."

Bryan Bischof on why truth-seeking is non-negotiable for data teams.

#DataScience #DataCulture

The most predictive feature in the world's first coffee recommender? Favorite salad dressing. The baristas knew.

Bryan Bischof on Data Renegades

#DataScience #MachineLearning

"When an escalator breaks, it just becomes stairs. When your data workload fails, it's often just stale data."

Scott Breitenother on why most data failures aren't as bad as they feel.

#DataEngineering #DataReliability

CL Kao takes the stage at the DataTune Conf tomorrow: "A Practical Playbook for Building Data Agents (That Don't Break Your Pipeline)"

288 benchmark trials. Three context problems every data agent hits.

CL and @doriwilson.com will be at the Recce booth all day. Come say hey.

#DataTune

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

Our own Kent Chen wrote up the architecture decisions and fixes the team landed on.

Read the full blog here: blog.reccehq.com/designing-re...

#dbt #DataEngineering #AI #BuildInPublic

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

The agent also produced wrong lineage graphs. Switched from dbt's parent_map format to explicit edge lists matching Mermaid's native format. Accurate ever since.

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

A lineage diff for 5 changed models blew past the 25k MCP limit. Fixes: dataframes over key-value, numeric indices over full node IDs, filter to changed + downstream nodes only.

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

Building Recce's AI Data Review meant working around three hard limits in Claude: 200k context window, ~90k single prompt, 25k per MCP tool response. 🧵

Wes McKinney built pandas in a mouse-infested NYC apartment on founder hours. Now he runs parallel Claude Code sessions and says AI is forcing "radical accountability" on every software vendor shipping mediocre products.

Full conversation: youtu.be/Uso8-yaERkE

#DataRenegades #pandas #ApacheArrow

"The data team can't afford to be two years behind. You need to get that exoskeleton going now."

Scott Breitenother on why full-stack data people using AI agents are the new operating model.

#DataEngineering #AIAgents

New engineers at Kilo ship code on day one. MVP feature by end of week. In production the next week. "We ruthlessly hunt down bureaucracy."

Scott Breitenother on Data Renegades

#AIAgents #Engineering

"These technologies, they're not robots replacing us. They're exoskeletons that make us better, faster, stronger."

Scott Breitenother on Data Renegades

#DataEngineering #AI

We think so too 😎 Have you had a chance to try us out @jeremynelson.bsky.social , would love to hear what you think about us in action!

Thank you @hivebox.bsky.social! If you haven't tried us, yet you should! Would love your feedback.

Couldn't agree more! Validation can't just live at the end. It needs to be woven into the plan an AI puts together from the start, which is why we built Recce and the Recce MCP. The tooling should be built for how teams are actually starting to work.

Scott Breitenother jumped into every Slack thread at Brooklyn Data. It made the company fast and him the bottleneck. His fix: subscribe to replies, don't comment, check back in 3 hours.

#Leadership #DataTeams

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

AI Data Review lands in the PR automatically. No manual queries.

blog.reccehq.com/designing-re...

#dbt #DataEngineering #DataValidation #AI #MCP

Designing Reliable AI Agents for dbt Data Reviews Code changes have AI review tools. Data changes don' - until now. Here's how we went from a single prompt to an AI agent that performs the first pass on data validation in every PR.

Code changes have AI review tools. Data changes don't... until now.

Our own Kent Chen wrote about how the team built a multi-agent system with Claude Agent SDK and MCP that reviews data changes in every dbt PR. Orchestrator + two specialist agents using 6 Recce MCP tools.

"The team still exists, but it's not five humans, it's one human and four agents."

Scott Breitenother on why the unit of productivity has changed.

#DataEngineering #AIAgents

Bryan Bischof's hot take: self-serve data analytics is the hype trend most likely to derail.

"People are selling third base. I'm still on first base."

#DataEngineering #Analytics

Absolutely agreed, the work won't disappear. More people will are now enabled to get hands-on in data workflows.

Full conversation here:
youtu.be/qKFBaDWMxkk

#DataEngineering #AIAgents #AICodingAgent

New episode of Data Renegades is live

Scott Breitenother built a 100-person data consultancy, watched himself become the bottleneck, and rebuilt everything at Kilo Code

Their data team is one person plus four AI agents.

"I shit you not. I found one where it gave three backpacks."

Bryan Bischof on the production bug where out-of-distribution items sat near the null vector and showed up in every recommendation.

#MachineLearning #DataScience

"Saying 'you are wrong' is not curious. Saying 'why are your priors different than what the data is showing' is curious."

Bryan Bischof on Data Renegades

#DataScience #DataCulture

Five Days, Five Data Problems, Five Fixes: What the Data Valentine Challenge Revealed Five companies tackled real data problems live. Agent benchmarks, DuckDB reconciliation, no-code pipelines, dbt cleanup, and data versioning in one week.

All sessions are live up on our youtube.

Or check out our blog summarizing the week: blog.reccehq.com/data-valenti...

#DataEngineering #DataValentineChallenge

Happy Valentine's Day to everyone who spent this week falling back in love with their data stack.

Five days. Five companies. No slides. No safety nets.

Thank you Greybeam, dltHub, Database Tycoon, & bauplan for doing this with us and shipping in front of an audience.

Data Renegades | Ep. #6, From Big Data to Curiosity-Driven Insight with Roger Magoulas | Heavybit On episode 6 of Data Renegades, CL Kao and Dori Wilson of Recce speak with Roger Magoulas about the real bottlenecks holding data organizations back.

"The best analytics asks more questions than it answers."

Roger Magoulas on why dashboards shipped is the wrong success metric. The real measure is whether the work generates questions worth investigating.

🎧 www.heavybit.com/library/podc...

#dataengineering #analytics

An AI Agent Built My Entire Data Pipeline. Here's How I Kept It From Breaking Production Aldrin from Bauplan closed the Data Valentine Challenge with a demo where he didn't write a single line of pipeline code. Claude Code did all of it — importing satellite telemetry into a lakehouse,…

Five days. Five companies. Five chances to fall back in love with your data stack. That was the Data Valentine Challenge

Full replays: youtu.be/yzX05Z8FlYw

#DataEngineering #DataValentineChallenge

An AI Agent Built My Entire Data Pipeline. Here's How I Kept It From Breaking Production Aldrin from Bauplan closed the Data Valentine Challenge with a demo where he didn't write a single line of pipeline code. Claude Code did all of it — importing satellite telemetry into a lakehouse,…

The prompting lesson: "It's better to explicitly say 'don't use Pandas' rather than just encouraging other libraries." Ban what you don't want. Don't just suggest what you prefer. Works for AI agents. Works for code reviews too.