01Design
Recovering an agent after a runtime failure
A recovery design for saving useful agent state, checking unfinished work, and controlling the return to operation.
4 min read
Read articleArticles
Notes on building and assessing AI systems
Short, practical notes on deciding where AI belongs, how to make it dependable, and when a simpler system is the better answer.
02Optimise
Choosing a quantisation for your local model
Budget runtime memory, compare candidate files, and test whether a smaller representation still supports the workload.
4 min read
Read article03Secure
Testing what an agent can reach and change
Use disposable fixtures, enforced access controls, and independent records to test an agent's capabilities.
4 min read
Read article04Build
Designing agents with clear permissions and approval points
Use an invoice workflow to define what an agent can do, when it needs approval, and how to handle mistakes.
3 min read
Read article05Transform
Taking AI from pilot to everyday work
Choose a workflow, redesign the work around it, and measure whether the change improves the operation.
3 min read
Read article06Evaluate
How to decide whether an AI system is ready for production
Turn workflow requirements into test cases, review failures, and make a release decision for a defined scope.
3 min read
Read article07Rationalise
Deciding when to reduce or replace AI
Review the result, count the full operating cost, and decide what to keep, narrow, migrate, replace, or retire.
3 min read
Read article08Transact
Controlling agent purchases from request to fulfilment
Follow a purchase through proposal, approval, payment, and delivery, with controls for limits and uncertain outcomes.
3 min read
Read article