Systems,
Engineering &
Truth.
We write about what actually works in production, not what looks good on a slide deck. Post-mortems, architecture reviews, and engineering philosophy.
Prompt injection isn't a model problem, it's an architecture problem
No system prompt stops a sufficiently motivated attacker. The fix isn't a better prompt, it's treating every tool an agent can call as a privilege boundary.
Agentic AI vs. workflow automation: what's actually different
Not every multi-step AI feature needs an agent. Here's how we decide between a deterministic DAG and genuine agentic reasoning, and why getting it wrong is expensive in both directions.
RAG vs. fine-tuning: how to actually decide
They get pitched as competitors. They're not. Here's the decision framework we actually use, and two engagements where the 'obvious' choice would have been wrong.
The hidden cost of LLM APIs: how we cut one client's inference bill by 80%
Hosted frontier APIs are the right call for a prototype and the wrong call at scale. Here's how we moved a client's highest-volume workload to self-hosted inference without touching accuracy.
Building platforms that developers actually want to use
Internal Developer Platforms (IDPs) fail when they are treated as gatekeepers. They succeed when they are treated as products.
The myth of 'Zero Downtime' migrations
Stop promising zero downtime. Embrace the error budget, communicate clearly, and engineer for rapid recovery instead.
Why we abandoned Kubernetes for raw EC2
A controversial but necessary move that saved our client $2M annually and reduced deployment complexity by an order of magnitude.
Stop guessing.
Start shipping.
Thirty minutes with a senior engineer. No sales deck, no junior SDRs. Just a technical conversation about what's actually breaking or slowing you down.