Mat Keep

I’ve had worse starts to the week. Somewhere on the Med between the coasts of Spain and Morocco

With some help from @cyera_io, we finally measured how broken permissions are osohq.com/post/we-finall…

What's not to love - #opensource and #AI. @RedHat turns to @ApacheAirflow and @astronomerio to deliver #agenticAI astronomer.io/blog/airflow-i…

The next data stack shift: Old world → #ETL pipelines → dashboards New world → #context pipelines → #AI agents Models aren’t the bottleneck anymore. Context is. The winners will build the AI context layer: substack.com/home/post/p-19…

If #AI #agents become the primary “users” of #databases, a lot changes: query patterns, isolation, ephemeral environments. Interesting industry perspective. Keen to see how other database vendors approach this shift. thenewstack.io/databases-for-…

Visa-scale data orchestration I put together a new Airflow in Action case study on how @Visa runs @ApacheAirflow for critical #Analytics and data workflows, with a ready made #Context layer for #AI astronomer.io/blog/airflow-i…

Feeling grateful for all of the kind and uplifting birthday messages I received this year…..well most of them anyway

If your #AIAgents can call tools, read data, or execute workflows…you’ve built an authorization problem. I wrote a guide on preventing over-permissioned agents for my friends over at @osoHQ : osohq.com/learn/how-to-p…

The team at @datadoghq processes 100+ trillion events per day. How do you orchestrate data workflows at that scale? In our latest @ApacheAirflow in Action case study, Datadog shares why they adopted #Airflow 3: astronomer.io/blog/airflow-i…

Seeing @ApacheAirflow increasingly deployed as a shared platform across #Engineering teams. Dag-level roles on @astronomerio add the missing piece: fine-grained access control so teams only see and operate the pipelines they own. astronomer.io/blog/dag-level…

How #SoftwareEngineering is evolving in face of #AgenticAI. Specs should be equations, not essays. Precision wins over prose when AI writes code. Also addresses risk of "cognitive debt" fromanengineersight.substack.com/p/specs-should…

Left #MongoDB thinking CAP theorem was behind me. #Agents pulled me right back in: Capability, Autonomy, Permissions: pick 2. The difference: CAP is physics. The agent version is an infrastructure problem and these are solvable - the @osoHQ take osohq.com/post/the-cap-t…

Nearly 50% of #AI #agentic usage today is in #softwareengineering. The other half? Fragmented across 16+ industries with most under 5% each. Coding may be saturated. Vertical AI agents are the wide-open frontier. garryslist.org/posts/half-the…

We made an implicit bargain with overpermissioning: bad, but not bad enough to fix. It worked because humans sleep, work business hours, slow down. Agents have no remorse and no speed limit. The bargain is over. osohq.com/post/least-pri…

"Cognitive debt" is gaining traction. Fueled by #vibecoding it slows teams down more than #techdebt, accumulating when assumptions aren’t documented and context is lost. Product Requirement and Design Decisions docs have never been more important! simonwillison.net/2026/Feb/15/co…

Always wanted to write a case study about #Oracle - and now I can! How they use @ApacheAirflow to orchestrate #AI pipelines across clusters of scarce #GPUs astronomer.io/blog/airflow-i…

Looks like #Claude Sonnet 4.6 is a serious step up in long-context reasoning, and agent workflows, especially with the 1M token window. Looking forward to trying it out to see how it holds up on real projects. thenewstack.io/claude-sonnet-…

Conjecture that #AIAgents may drive new #programming languages designed for clarity and agent reasoning. With code cost falling, languages with explicit structure and simpler tooling could outcompete old ones. lucumr.pocoo.org/2026/2/9/a-lan…

Team at #Microsoft warns of “#AI Recommendation Poisoning”: hidden prompt injection in AI summarize, skewing future recommendations without users knowing. Security must evolve with #agents. This is why I'm excited to be working with @osoHQ microsoft.com/en-us/security…

Interesting post - #Java is winning production #AI: 62% of enterprises now run AI apps on Java. #Python still dominates research and prototyping, but the claim is Java leads where scale, security, and uptime matter: thenewstack.io/2026-java-ai-a…