Writing & research
We publish the working out, not just the conclusions.
Over 243 hands-on essays on LLMs, agents, RAG and the business of AI — most written while the thing was still breaking. The research desk behind the advisory work is the same one behind the writing.
Where to read
Three places, three purposes. The essays are the thinking; AI Insights is the searchable archive; Director's Cut is the weekly summary for people who don't have time for either.
Medium
Long-form essays on AI practice: model releases tested rather than summarised, architecture walkthroughs, and post-mortems of builds that didn't work. Published in Stackademic and Python in Plain English among others.
AI Insights
A working knowledge platform built around 243+ essays: semantic Q&A with citations and dates, text and voice input, playbooks, bookmarks, topic exploration, quizzes, related reading and a business-triage mode.
Director's Cut
The weekly digest — what shipped in AI, notable reading, what we built that week, cinema picks, and the threads worth pulling between them. Email-client-safe, tracked links, one sitting.
Themes from the writing
A sample of the questions the archive keeps returning to. Follow through to Medium for the current pieces.
Model testing
Ten tests against a top-two claim
An image model claimed a top-two arena ranking and a set of new editing tools. We ran the claims one at a time rather than repeating the announcement.
Tested, not summarised.
Security
The Trojan Horse was a swarm
A published postmortem revealed an incident was bigger and stranger than the first account suggested. What it means for anyone running agents near production systems.
Follows the correction, not just the story.
Evaluation
Queeg's Strawberries
An evaluation layer built for an image pipeline caught nothing across 186 real images — and finding out why took six broken attempts. A post-mortem of our own work.
A failure, published.
Consulting
Inside the AI side of a consulting project
How bounded sub-agents were used on a live engagement — and why orchestration mattered considerably more than agent count.
From an engagement, written up.
Strategy
Changing the track layout
The infrastructure question underneath the model race, and why the visible leaderboard may be measuring the wrong thing.
Strategy lens. A recurring question in the archive.
Practice
The Thinking Loop
How the build, the writing and the advisory feed each other — the working method behind everything on this site.
Practice note. How building, writing and advisory reinforce each other.
Start a conversation
Tell us what's actually breaking.
Not what you think the AI solution is — that part is usually wrong, including when we guess it. Describe the problem and you'll get an honest read on whether it's worth building.
First conversations are with Lakshmi Narayana U, Founder & Principal Consultant.