Insights
Notes from the field.
Practical writing from our engineers — short, opinionated, and grounded in production work. Temporal, workflow automation, agentic systems, generative AI, and the parts of software that don't make the conference talks.
All notes
Designing durable agent workflows for production
Production agents that survive retries, partial failures, and weeks-long executions don't look like a chat loop. They look like a workflow with an LLM in the middle.
Trillion Thoughts Engineering · 2 April 2026
Eval-driven development for LLM applications
If you don't have evals, you don't have a product — you have a vibe. Here's the playbook we use to make LLM behaviour testable, repeatable, and shippable.
Trillion Thoughts Engineering · 22 March 2026
Prompt caching at production scale
Prompt caching can drop your inference cost by an order of magnitude — but only if you structure prompts the way the cache wants you to.
Trillion Thoughts Engineering · 8 March 2026
Workflow automation isn't a no-code problem
The pitch is always 'drag boxes, ship workflows.' The reality is that durable, debuggable, multi-week processes need real engineering — even when the surface is visual.
Trillion Thoughts Engineering · 26 February 2026
From monolith to modular: the cuts that actually pay off
Most monolith decompositions fail at boundary selection. Here's the heuristic we use to find the cuts worth making — and the ones that look attractive but cost more than they return.
Trillion Thoughts Engineering · 12 February 2026
Saga patterns in Temporal
Distributed transactions don't go away when you split the monolith. Sagas — and Temporal's compensation model — are how we keep them sane in production.
Trillion Thoughts Engineering · 29 January 2026