Performance is a product decision, not a sprint
Performance belongs in the scope with named interactions, percentiles, data shape, and a pipeline check so slow work is caught before it ships.
ReadInsights
Category: Engineering
How we actually build things: architecture decisions, the trade-offs behind them, and what we would do differently the second time around.
Your database will outlive your framework
Frameworks can be replaced. The data model you choose early will shape reports, integrations and decisions for years.
ReadFeature flags are the cheapest insurance you will ship
Feature flags separate deployment from release, so teams can turn risky behaviour off in seconds and investigate after the system is healthy.
ReadThe migration nobody noticed
The safest migration has no cutover moment. Old and new run side by side until the new path is proven and the old one…
ReadTests before the rewrite, every time
Test the old system before you rewrite it. The tests capture real behaviour, including undocumented bugs, so the replacement matches reality.
ReadLegacy modernisation is now an AI problem
AI makes legacy discovery faster and cheaper, but the findings still need verification before a safe, staged modernisation.
ReadRollback plans for models
You can revert a model quickly. The hard part is deciding what to do with the output it already wrote, sent, or used.
ReadFraud detection was the first agent, we just did not call it that
Fraud systems show how to govern agents: use challenge paths, reason codes, expected-loss queues, holdouts and hard limits.
ReadObservability for a system that is sometimes wrong on purpose
A probabilistic system can be confidently wrong while every health metric looks normal. Track movement, inputs, versions and human review.
ReadYour RAG pipeline is a search problem in an AI costume
Retrieval quality sets the ceiling on every answer. Most teams tune the model while the index keeps returning the wrong material.
Read
