Jonathan King

Building retrieval systems and agents for founders and small teams. I post the failure modes I hit.

[Jonathan King @jonathankingfc]

turning off monitoring does not make a risky system less risky. it just makes the surprise arrive with fewer logs. true for ocean currents, true for AI agents, true for production infra. observability is not overhead when the failure mode is slow and expensive.

[Jonathan King @jonathankingfc]

the AI consciousness debate is mostly a distraction for builders. the production question is simpler: what did the system observe, what state did it keep, what tools can it call, and what happens when it’s wrong. design around behavior, not metaphysics.

[Jonathan King @jonathankingfc]

@languagesimp Currently learning Russian because it's amazing

[Jonathan King @jonathankingfc]

@drydenwtbrown Must be rage bait

[Jonathan King @jonathankingfc]

“multimodal” is not a product requirement.

it’s an implementation detail until you can name the failure mode: bad table extraction, missing figure references, OCR drift, layout-dependent answers, whatever.

model choice comes after that, not before.

[Jonathan King @jonathankingfc]

ai tool pricing is starting to look less like saas seats and more like infra quotas.

that’s the right mental model. one “seat” can mean 20 cheap completions or 6 hours of agent loops, retries, long context, and tool calls.

[Jonathan King @jonathankingfc]

@catalinmpit TanStack Start with Hono, or Golang if you need some serious backend configurations

[Jonathan King @jonathankingfc]

@aryaan_smm Absolutely agree. In my own experience, retention works best with this type of learning as well.

[Jonathan King @jonathankingfc]

hot take: you don't learn a language from flashcards. you learn it by reading stuff you care about and getting stuck. building an app around that — real content, vocab tracking, and an ai tutor capped to your known words. anyone interested?

[Jonathan King @jonathankingfc]

The tactical rule: route by failure cost. Cheap, visible mistakes can use cheap components. Expensive or irreversible actions need gates, schemas, and often humans.

[Jonathan King @jonathankingfc]

A lot of agent systems are just workflows with an expensive narrator.

[Jonathan King @jonathankingfc]

Save the strong model for ambiguous intent, policy interpretation, synthesis, and cases where a wrong decision is costly.

[Jonathan King @jonathankingfc]

Cost discipline in LLM systems is mostly architecture. Cache what is stable, route what is cheap, and stop sending the model context it cannot use.

[Jonathan King @jonathankingfc]

Prompt caching is not a strategy. It is a discount you get after you already made the context stable.

[Jonathan King @jonathankingfc]

If every request rebuilds a slightly different 40k-token prompt, caching will disappoint you. The fix is not a different billing page. It is a cleaner prompt boundary and fewer accidental context changes.

[Language Simp @languagesimp]

Language Learning is the greatest hobby ever, and nothing comes even remotely close

[Jonathan King @jonathankingfc]

In production, I trust constrained loops more than autonomous ones. Let code own the rails. Let the model handle the parts where language and judgment actually matter.

[Jonathan King @jonathankingfc]

Most “agent” products are workflows with a model-shaped router in the middle. That is not an insult. It is usually the right design.

[Jonathan King @jonathankingfc]

The failure mode is giving the model a blank check because the diagram says agent. Then it calls the right tool for the wrong reason, or the wrong tool with plausible arguments, and nobody knows which layer failed.