Posts tagged "ai-agents"
-
Friday Roundup #32: The Compliance Gap
A federal judge voided a Flock search, a court ordered the Bureau of Prisons to restore union rights, and California subpoenaed OpenAI — then the Bureau terminated the contracts again, so stop recording the date a control was ordered and start recording the date somebody performed it, and at whose expense.
-
Friday Roundup #31: The Hand on the Pen
Everybody wrote something down this week — a California AI labor statute, a White House superintelligence accord, Flock's answer to Congress, OpenAI's own training pause — and the only paper that binds is the paper whose author is not the party being bound, so for every control in your runbook, name the hand that held the pen.
-
Friday Roundup #30: The Enforcement Nobody Wrote
Nothing was stopped by a rule this week — Flock is dying of cancelled contracts, nearly 90% of Amazon's agent prototypes died in engineering, and a court order got probed for holes the next morning — so if you need something to actually stop, attach a cost to it and name who sends the invoice.
-
The Seven Kinds of Agent Memory (and Which Ones We Actually Have)
There isn't one agent memory — there are at least seven. We audited our fleet against the taxonomy and found we were strong where it's cheap and thin where it counts.
-
Context Engineering: Your Agent's Memory Is a Junk Drawer
We audited our AI agent's memory and found 56% of its search index was redundant voice transcripts. Here's the three-tier architecture we designed to fix it.