Technical notes on production AI: long-context memory and browser-agent token efficiency.https://synthworx.com/en-ushttps://synthworx.com/notes/seven-production-failures/https://synthworx.com/notes/seven-production-failures/Seven repeatable failure modes: the half-working vendor tool, the manual document workflow, classification at scale, the agent that stops on record 41, the runaway bill, the expensive wrong answer, and the unowned roadmap.Thu, 05 Nov 2026 00:00:00 GMThttps://synthworx.com/notes/what-a-diagnostic-finds/https://synthworx.com/notes/what-a-diagnostic-finds/What the two-week AI Production Diagnostic contains, what it typically finds, and why it comes before a build. Includes the cost model, the eval gap, and the go/no-go.Mon, 02 Nov 2026 00:00:00 GMThttps://synthworx.com/notes/tappi-token-efficiency/https://synthworx.com/notes/tappi-token-efficiency/How Tappi Browser reduces browser-agent token usage by indexing interactive elements, returning bounded observations, and keeping the action surface typed. Includes the token arithmetic and the failure modes the design removes.Wed, 28 Oct 2026 00:00:00 GMThttps://synthworx.com/notes/long-context-memory/https://synthworx.com/notes/long-context-memory/A technical note on Kamaloki: writing a session's KV cache to disk, probing it for relevant spans, and splicing them back into a small model's context. Includes the measured results and the parts that did not work.Tue, 20 Oct 2026 00:00:00 GMT