Raj

engineer who writes about AI agents, legacy systems, and the gap between conference slides and production.

[Raj @Raj_tldr]

Two talks at @code_europe , done. Warsaw is incredible. Best organised conference I've been to this year. Thanks Marta and team.

Next talk: NDC Oslo on Friday.

[Raj @Raj_tldr]

Two of my favourite topics at @code_europe next week.

1. Forensic DDD — reverse-engineering domains from legacy artifacts
2. Beyond the Codebase — giving coding agents the business context they're missing

Warsaw, Sep 14 & 15.

[Raj @Raj_tldr]

I first shared this as a “Demand-Driven Context�� talk at @aiDotEngineer and @CreateWithHQ

Context Blocks is the OSS version: typed KB + multi-persona eval + CLEAN / INCOMPLETE / MISSING / TRIBAL.

MIT. pip install context-blocks
Repo: https://t.co/qftYELwfcc

[Raj @Raj_tldr]

Only ~20% of what my AI agents needed was actually documented. The rest was partial docs, tickets, Slack threads, or pure tribal knowledge.

That pattern kept breaking real projects, so I built Context Blocks to measure that gap.

[Raj @Raj_tldr]

Spoke at @CreateWithHQ in Brighton last week.

Your AI agent doesn't have a model problem. It has a knowledge problem.

Context Blocks is now open source — it maps exactly where your knowledge breaks down.

[Raj @Raj_tldr]

Speaking at @Createwith AI builders conference in Brighton next week.

AI agents are getting scary capable. But most of the knowledge they need to actually be useful? Not written down anywhere.

That's my talk. Mapping Domain Complexity for AI.

[Raj @Raj_tldr]

How do 1.4 billion Indians learn multiple languages without textbooks? From the streets. Observe, infer, validate.

Same method works for rediscovering domains from legacy artifacts. Some people called it "Street DDD." I'll take it.

Next one's in Amsterdam. Home turf.

[Raj @Raj_tldr]

One week out from @ddd_eu in Antwerp. Talking about what happens when you inherit a legacy domain and the textbook doesn't apply — 7 forensic techniques for discovering domains from production artifacts instead of sticky notes.

[Raj @Raj_tldr]

Two weeks. Antwerp. The system remembers everything. We're the ones who forgot. See you at DDD Europe, June 10th.

[Raj @Raj_tldr]

Interesting pattern building an enterprise context engine: the agent never asked for code docs, sprint artifacts, or org charts. It kept asking for system relationships, past incidents, and why decisions were made. We document what's easy to write. Agents need what's hard to find

[Raj @Raj_tldr]

@ashwingop The "expensive amnesia" framing is spot on. But the harder problem isn't giving agents memory — it's knowing what's worth remembering. Most enterprise knowledge lives in people's heads, not in any system an agent can reach.

[Raj @Raj_tldr]

Building an enterprise context engine. Instead of stuffing agents with docs, let them tell you what they're missing. Currently recording every time the agent confidently gets something wrong. The footage is... entertaining.

[Raj @Raj_tldr]

52% of the knowledge our AI agents needed didn't exist in any document. Not buried deep. Not hard to find. Just never written down. That's the real AI-readiness problem.

[Raj @Raj_tldr]

@thekitze I work at IKEA and I can confirm step 7 always has leftover screws 😄

[Raj @Raj_tldr]

@swyx Happy 40th! This hit home — especially "study everybody's game, define your own, then play it." Great meeting you in London. Here's to the next decade.

[Raj @Raj_tldr]

Back on X after a long detox. Ran a workshop at @aidotengineer Europe — Demand-Driven Context: let AI agents fail and tell you what they're missing. TDD for agent knowledge. Packed room, great conversations.

Paper: https://t.co/jE6qeotv6L
slides: https://t.co/5DAf2Vc4JN