Not long ago, reducing a sprawling workflow to three taps felt close to the endgame. Then even the three taps disappeared.
The “three-tap guy” became the most ironic victim of process improvement: he optimized the job so well that BPR eventually eliminated his own position. The routine of opening a separate computer, asking Codex to edit, checking the result, visiting GitHub, switching contexts, and thinking about extra usage costs largely vanished as well.
What remains on the human side is almost comically small: talk with AI, research and think, decide that a discussion is worth preserving, summon a saved instruction with roughly two typed characters, and send it. Article production, 12-language localization, quality checks, repository updates, and publishing continue behind the scenes.
From the outside, this can look like impulsive same-day action. It is almost the opposite. Much of the thinking, comparison, and rule-making has already happened. Execution looks fast because the decision cost was prepaid.
1. When three taps felt like the endgame
Compressing a workflow with many manual steps into three taps is already serious process improvement. But improvement changes the landscape: once the big sources of friction disappear, the three tiny taps that used to be irrelevant become the biggest remaining friction.
That creates the next question: “Why does a person need to press these three things at all?” Yesterday’s hero becomes tomorrow’s redundancy candidate.
2. Then the three-tap guy got fired
The first stage of automation makes work faster. The next reduces the work. A later stage removes the reason the work exists at all.
The three-tap guy was not fired because he was lazy. He was almost too competent: by standardizing his job to the limit, he created the conditions under which the system no longer needed him.
3. Codex and GitHub left the foreground too
The old process still required opening a computer, asking Codex for edits, inspecting results, correcting them, visiting GitHub, and checking state. None was huge, but every tool and context switch added activation energy.
Now Codex looks less like a full-time operator and more like an on-call specialist for incidents or major reconstruction. GitHub did not become unnecessary; it became background infrastructure. When a person stops opening GitHub every day, GitHub has finally become a proper backend.
4. The current job is “conversation → two characters → done”
The primary interface is conversation. Talk with AI, investigate questions, consult public sources or research when useful, test objections, change the angle, and clarify a thought.
When something deserves to become durable rather than disappear as chat history, roughly two characters call up the saved instruction. Send it, and the human part is largely over.
Even the “template-copy guy” has become an honorary employee who works for two keystrokes. Those keystrokes are no longer writing labor. They are an editorial approval gate: the human decides what is worth turning into an asset; the production line executes.
5. When AI starts eliminating AI work
Human work is delegated to AI. The request to AI becomes standardized. The standardized request is triggered automatically. AI handles exceptions. Some exceptions become rules.
AI: “What is my assignment today?”
System: “Nothing, unless something breaks.”
This is not the technological singularity in the strict sense. It is a small form of recursive or second-order automation: once one layer is automated, operating that automated layer becomes the next automation target.
6. Why this kind of system is easier to sustain
Consistency is not determined only by minutes. Opening a computer, switching services, remembering a procedure, checking output, and switching back are tiny entrance fees paid every time work begins.
If useful knowledge already emerges from ordinary conversation and a two-character trigger routes it into production, there is almost no separate “write an article” session to initiate.
The by-products of ordinary thinking and research become content inventory. The deepest benefit of automation is not merely minutes saved; it reduces the opportunities to say, “This is annoying; I’ll do it another day.”
7. What remains for the CEO is value judgment
As operations disappear, the human role becomes strangely executive. There is less need to touch files, coordinate translations, or press publish.
What remains is deciding what is interesting, what deserves preservation, what quality bar matters, and what should improve next.
“That is worth keeping.” Two characters. “Turn it into an asset.” The factory behind the wall starts moving.
An article is not a financial instrument and does not guarantee revenue. But turning fleeting discussion or investigation into searchable, reusable, updateable material creates accumulating digital content inventory with potential future value.
8. Improvement ability and process design are the durable skills
As AI gets better, “knowing how to operate this AI tool” becomes a fragile moat because the operating procedure itself can be automated.
More durable skills sit upstream: finding waste, defining objectives, setting quality criteria, classifying exceptions, deciding what deserves automation, deciding where human approval must remain, and finally asking, “Does this step need to exist at all?”
If AI becomes 100 times faster while executing a pointless workflow, it simply produces waste 100 times faster. The stronger capability is to redesign the work itself.
The strongest process improvement is not making a task faster. It is removing the reason the task exists.
9. Using P-like exploration and J-like standardization together
MBTI is only a playful metaphor here, not a diagnosis.
Exploration asks, “Is there a better way?”, “What other angle exists?”, “Can we change the premise?” Operations need another skill: “Use this rule from now on”, “These are the conditions”, “Handle this exception this way”.
The powerful loop is: explore → discover a better method → freeze it into a rule → automate it → question whether the frozen step is still necessary. Explore like P, stabilize like J, then come back and break the stabilization like P again.
10. Same-day execution is not impulsivity
Visible speed can be misleading. The subject may have been considered for a long time. Information was gathered. Alternatives were compared. Failure conditions were considered. Decision criteria were built.
Once the threshold is crossed, there is simply less left to debate.
Deliberate → structure → pre-build decision criteria → cross the threshold → implement the same day. Observers see only the last step. The decision cost was paid earlier.
11. “Ultimate” keeps getting revised
Three taps genuinely felt close to ultimate at the time, and relative to the old system, they were excellent. Process improvement simply has local endgames.
100 steps → 10: “Ultimate!”
10 → 3: “Now this is really ultimate!”
3 → 1: “There cannot be anything left!”
1 → “Do we need that one step?”
When the biggest source of friction disappears, a smaller one becomes the new bottleneck. Every surpassed “ultimate” is evidence that visibility improved.
12. Should the final two characters disappear too?
Not necessarily. If they were purely mechanical, deleting them would make sense. But they also confirm an editorial decision: this particular conversation is worth preserving as an article.
Remove that gate and the system begins deciding what deserves publication by itself. Volume may rise, but so can noise.
The goal is zero unnecessary labor, not zero human agency. The final two characters are closer to a signature than a job.
13. Conclusion: delete work instead of merely accelerating it
Large manual workflows became three taps. The three taps disappeared. Routine Codex operation disappeared. GitHub moved out of the foreground. Even the template command shrank to roughly two characters.
What remained on the human side was upstream work: conversation, deliberation, finding angles, design, value judgment, and improvement.
The speed did not come from thinking less. It came from thinking earlier, preserving good decisions as rules, and refusing to pay the same decision cost again.
Find an angle → deliberate → design → act → observe → improve → standardize → question the standard itself.
The three-tap guy is gone. Codex is close to on-call status. The template-copy guy clocks in for about two characters. And the CEO keeps talking with AI while the factory quietly adds inventory in the background.
