Pillar 01
Agentic systems
Topology, tool surfaces, memory, evaluation, observability — agents that survive a real workload.
§ Insights
Writing for senior practitioners.
Articles that pass the test of "would a senior engineer respect this?" — covering the five pillars below. We publish when a piece is worth reading, not on a cadence calendar.
§ Agentic systems ·
Evals for agents that use tools, not just tokens.
Most agent evals score the final answer and stop there. That's a demo eval, not a production eval. This piece walks through how we build eval spines that score trajectories — tool calls, sub-goals, cost budgets, rollback behaviour — and why that changes what you can safely ship.
§ Agentic systems ·
Three ways to bound an agent loop before it eats your budget.
Every agent programme discovers, usually painfully, that "the loop won't terminate" is the default behaviour of loops. Here are three specific bounding patterns we reach for — each with the failure mode it's actually preventing.
§ LLM integration ·
RAG for the meeting-notes-hostage: retrieval when the corpus is a mess.
Most RAG failures are actually retrieval failures dressed up as generation problems. This piece walks through what "retrieval on a real corpus" looks like — chunking that respects structure, hybrid retrieval, re-ranking that pays for itself, and the eval spine that keeps you honest.
§ LLM integration ·
We measured our prompts. Here's what changed.
Six months of "prompts live in git, measured on every merge, versioned like configuration". A short field note on what the discipline actually caught, what it missed, and what the team wouldn't give back.
§ Multi-modal AI ·
Speech-in, structured-out — a small pattern that scales.
A small, boring, reliable pattern for turning a voice interaction into a structured record. Three stages, one schema, one repair loop, and a measurement bar that catches drift before users do.
§ AI strategy & leadership ·
What "AI transformation" actually costs the second year.
Year One is the pilots. Year Two is the bill. This piece walks through the five cost lines that show up in Year Two of every AI programme — none of them are the ones the Year One business case forecast — and what to do about it before the invoice arrives.
§ AI strategy & leadership ·
The board deck slide we keep having to write.
A short field note on the one slide that keeps carrying every AI-strategy board conversation we've been in. Not because the slide is clever — but because the frame it sets makes the rest of the deck redundant.
§ AI-aided development ·
A senior engineer's checklist before letting an agent commit.
A concrete checklist for the moment a senior engineer is about to approve an agent-drafted change. Eight items. No slogans. What we ask ourselves before pressing merge.
The pillars
Five pillars we write under.
Every article belongs to exactly one pillar. Secondary tags handle everything else — model, vendor, pattern, audience.
Pillar 02
LLM integration
Chat, RAG, chunking, prompt and eval management — the integration stack and the metrics that catch regressions.
Pillar 03
Multi-modal AI
Vision, speech, video, image — orchestrated across vendors with quality and cost monitoring you can live with.
Pillar 04
AI strategy & leadership
For leaders who must call consequential decisions without full certainty — portfolios, operating models, governance.
Pillar 05
AI-aided development
AI coding agents in the loop — tooling, review gates, safety rails, and the measurement that makes adoption sticky.
Pillar 06
Workflow automation
RPA, workflow orchestration, and where AI-in-the-loop earns its keep — the boring engineering that turns operations from manual to measurable.
Want to know when new pieces land?
We won't pad the list with anything that wouldn't pass our own review. Subscribe and you'll hear when something ships.