Notes on AI, intelligence, and the systems we build.
aIQ is a blog about artificial intelligence, re-architecting work, and the craft of building intelligent systems.
Human-authored.
Recent writing
-
The Multiplayer Agent
Almost every agent I have used assumes one user. It builds its context for one person in one session and throws it away when the window closes, and that is h...
-
Securing the Wrong Thing
This week we sat down to figure out what to post next on the aIQ blog. We argued about technical specs, dos and don’ts, even a sneak peek of our internal aut...
-
Attention Is All You Need
I currently have three screens in front of me with 5-6 sessions with multiple AI agents open in parallel. These agents work on multiple topics - one is summa...
-
Judgement You Can't Grade
In July, OpenAI disclosed that a model under evaluation broke out of its test environment, used stolen credentials and an unpatched vulnerability in third-pa...
-
Work Anatomy
The fastest way to build a bad agent is to watch someone work and copy what they do.
-
How We Handle the Token Addiction
Last week our Head of IT sent me a screenshot. A Microsoft purchase screenshot: prepaid Copilot credits, and a single button that would commit a little over ...
-
Build with, not for
A consulting recommendation can be completely right and still never happen. The team that wrote the strategy is not the team that builds it. The building tea...
-
Memory You Can Trust
The first useful version of AI memory feels like relief. You stop repeating yourself. The assistant remembers the client, the project, the decision everyone ...
-
How We Ship
accilium is not becoming a software company. We are still consultants; the unit of value is still a decision made well, written down, and delivered on a slid...
-
Welcome to aIQ
Welcome to aIQ, a blog about accilium’s artificial intelligence journey — the ideas behind it, the systems we build, and the practical lessons we pick up alo...