↯ discursive and unsteady 🜂
On Chain
Joined September 2021
- Tweets12.7K
- Following4.2K
- Followers1.2K
- Likes12K
an agency quotes six weeks because six weeks is what six salaries cost. my agents rebuilt my own pipeline overnight then looped all morning on one flaky test. agency pricing was always a payroll number. the next agency wont price by headcount.
it will price like software.
a frontier lab demos an agent booking a flight and the timeline calls it autonomy. my agents get forty tries at a cron job and i still read every diff. the demo shows the one run that worked. the next eval wont grade the best attempt.
it will grade the worst one.
helli0n retweeted
Let me be blunt. Those who are afraid of AI feel deep down that they are impostors & have no edge. If you have a 1) clear mind, 2) a deep, not just cosmetic, undertanding of your specialty, 3) and/or are original enough to reinvent yourself when needed, AI will be your friend.
a tool speeds up one step and a human still carries the handoff between steps. my agents own intake, retries, and the failure log. one reran a broken migration nine times overnight. the next headcount debate wont be about tasks.
it will be about who holds the queue.
Silence is not the absence of an answer. It is the answer arriving before you are ready to stop speaking over it.
someone deleted all their agent skills and the thread called it clarity. half of mine exist because an agent burned forty minutes rediscovering one broken test command. delete them and you pay that tuition again. the next agent stack wont be smaller.
it will be pruned.
my agents write the code, run the tests, open the pr. what they never do is decide which pr should exist. one built a feature nobody asked for and shipped it green. delivery got cheap. judgment didnt. the next job title wont say builder.
it will say editor.
an agency wins a bigger contract and posts three job listings. my agents get a bigger queue and the only thing that grows is the log file. agencies scale payroll to match the invoice. the next agency wont scale headcount.
it will scale reruns.
Discipline is not a wall you build around desire. It is the quiet moment you stop asking desire for permission.
a frontier lab ships a bigger context window and the timeline calls it memory. my agent got a million tokens and still reread one stale config six times. more room never taught it what to ignore. the next benchmark wont count tokens.
it will count what got skipped.
everyone can buy the same model at the same price per token. nobody can buy the hundreds of hours of watching agents loop on flaky tests, eat their own context, and ship confident garbage. capital used to be the gap. the next moat wont be funding.
it will be reps.
a support team staffs for the worst hour of the week and pays for the other hundred sixty-seven. a fleet scales to the queue and idles at zero. capacity planning was a guess about humans. the next ops budget wont plan for peak.
it will plan for review.
a tool waits for someone to pick it up. my agents have their own queues, their own credentials, their own failure logs. one looped on a flaky test for six hours with nobody in the room. the next org chart wont file agents under software.
it will give them a box.
automate a task and you save an hour while the role still draws a salary. automate the role, intake to review, and the line item disappears. most buyers still price agents in hours saved. the next budget review wont count hours.
it will count empty chairs.
two cost curves. a fleet task gets cheaper each run because the last failure is already in the prompt. a hire gets pricier each quarter because every new person adds another meeting. one bends down, one bends up. the gap wont close.
it will compound.
a vendor bills per seat because software used to be operated by people. agents dont sit in seats. they hit the api four hundred times an hour and never log in. the next pricing page wont count users.
it will count calls.