Every tiffin so far
Browse by track: ISIntent & specs CKContext & knowledge ABAgents that build UMUnderstanding & modernisation VTVerification & trust POPeople & operating model
Your Agent Is a Great Sprinter and a Terrible Relay Runner: 95% alone, 51% in a chain
Agents finish 95% of tickets alone and 51% of twenty-ticket chains. Plus: PHP files that LLMs broke and a rule-based tool left alone, a story format agents can't misread, why trimming your prompt can raise your bill, and a 90% merge rate that proves nothing.
Your AGENTS.md Is a Stale Spec With a Loudspeaker: why your agent obeys the file over the code
Your agent believes the instruction file over your code, 90.1% of the time. Plus: 718 firms where pull requests rose and finished work didn't, a five-question AI budget review, a time-out for coding agents, and a COBOL number you can't check.
Slop Is Whatever Passes: the AI junk worth worrying about isn't the kind you can spot
Slop in code arrives with a green tick. Plus: the 91% review bill, the lookalike skill that took the job, a fifteen-minute decision pin, and why a made-up package is only harmless until someone registers it.
Pack a Tiffin, Not the Kitchen: your context file is billing you every turn
More context, same answers, bigger bill. Plus: AI apps that retire without warning, a ten-minute ticket lint, and the UPI rule every agent that retries should follow.