The argument

Anyone can build anything now. Knowing what to build is the job.

For thirty years the scarce thing in software was hands: who could actually write it. That constraint is gone, and it did not take the problem with it. What sits in its place is harder to buy. What is worth building, in what order, to what standard, and how you know when it is genuinely finished.

The second-order effect is the part most people miss. Building fifty times faster does not shrink the work, it multiplies the surface competing for a fixed amount of attention. More capacity makes selection and sequencing more decisive, not less. Everyone gets the capacity. The gap between capacity and judgment is now the whole game.

What the gap is made of

Five things capacity does not give you.

Each one below is a way projects fail now that they did not fail before, the scar it left on mine, and the instrument I built so it cannot happen the same way twice. The instruments are open source. The links go to the thing itself, not a description of it.

  1. SelectionDeciding what is actually worth building right now, before anything gets built.A public roadmap whose Q3 targets are a year in the past while the rows underneath them read Completed. A launch date named in April that is still ahead of you in September. Neither of those is a building problem. Every one of those items shipped. They were the wrong items, in the wrong order, and the capacity to build them quickly is exactly what made that possible.
    What now stops itskills/lamps

    Ten questions asked at four moments in a task, chosen because they are the ones the request did not contain. The expensive builds are rarely botched. They are flawless executions of a brief nobody examined.

  2. StructureArchitecture that still holds at the third feature, and at the third person touching it.I once made a four-line change that produced a 1,371-line diff, because the tool reformatted a registry file on its way through. In a repo where several agents hold worktrees at the same time, gratuitous reformatting is not untidy. It is other people's merge conflicts, and it hid the four lines that mattered.
    What now stops itdocs/ENGINEERING_PRINCIPLES.md

    Every rule written as the thing that went wrong first, with the month it cost something and the mechanism that now stops it. A principle without its scar is advice. With it, it is a decision you can hand to someone else.

  3. The last mileThe unglamorous distance between working in the preview and working for a stranger.On September 2nd a migration locked a table to the service role while the application was still writing to it with the anon key. Every call to the AI receptionist returned a 500 for hours. I found out by calling my own phone line. The check that would have caught it in sixty seconds already existed. It was scheduled to run the following Monday.
    What now stops itskills/clarity-gate

    The stranger's first five seconds, simulated before the thing ships, on the half a build gate cannot measure. A weekly health check is not monitoring, and a page you wrote is not a page anyone can read.

  4. Not drowning in slop loopsNoticing when the model is confidently going in circles, and stopping it.Three sessions in a row on the same bug. Each one ends with a fluent, structured, entirely confident summary of what was fixed. Nothing observable changed. The tell is that the summary describes the edits and never a re-measurement, and the model's own account of its failure is the least reliable thing in the transcript, because the process that wrote the bug also wrote the explanation.
    What now stops itskills/hedge-audit

    Find where caution has been quietly written into the code as a stub, a refusal, or a fabricated success. Paired with philosophy/, which names the tilt behind the reflex instead of treating each instance as a surprise.

  5. VerificationThe difference between it ran and it works, held to consistently.Seven ways I have watched a check lie. A 200 from a modern web route is the shell, not the page, so a soft-404 answers 200 all day. A piped exit code belongs to the last stage of the pipe, so a failed test suite reports success and the runner calls it green. A configured environment variable is a name, not a value: one of mine was listed as set, was an empty string, and the code's own guard failed silently into a log nobody read.
    What now stops itskills/definition-of-done

    A four-rung ladder: built, deployed, works, fed. You say which rung you have actually proven and what the proof was. Its companion, the seven cheap lies, is the list above, so the same check cannot fool the next person.

Check it yourself

None of that is a claim. It installs.

The whole set runs inside your own Claude, on your own project, in one command. It will start correcting you tonight, without me in the room. That is the honest test of whether any of this is real.

/plugin marketplace add Traviseric/best-practices
/plugin install best-practices@traviseric

What is deliberately not in it: the overnight autonomous runner, the goal-sheet system that decides what agents work on next, fleet-scale worktree discipline, and the Business Brain. Those need the operating system around them to transfer, so they live in the course and in the work itself. The free tier is the part that survives without me.

Where this goes

Three depths. Start at the free one.

01 / Free, nowThe kit

The instruments above, installed into your agent. Yours, with no conversation required. Most people never need anything past this.

02 / TaughtThe course

The operating system the instruments came out of, run end to end on a real project: verifiers, concurrency, overnight runs, and how a brief gets chosen.

03 / InstalledThe work

Judgment applied to your particular thing. Which of your fifty possible builds is next, structured so it holds, and finished to the rung where a stranger can use it.

The third one is the part that does not compress into a repository, because it is not a method, it is a decision about your business. If you have the capacity and suspect it is going into the wrong things, that is the conversation worth having.