For a while, the internet was gone.
Then the connection came back.
I checked the computer at home and found something slightly absurd: development had kept moving.
The person was traveling through the sky. The PC was still at home. Meanwhile, one AI role was organizing documents and direction while another was actually working on the code.
The immediate reaction was simple:
“Wait. Do you even need me anymore?”
That feeling makes sense. But it is only half true.
What is disappearing first is not the human. It is the old rule that development stops unless the human is sitting in front of the keyboard and manually feeding in every next step.
1. Why does this feel so different?
Traditional solo development is tightly tied to presence.
You sit at the computer. You write code. You research. You fix something. You test it. You fix it again.
When you leave, the work usually leaves with you.
Agent-based development breaks that link.
The human can be traveling. The network can disappear for a while. Tasks that were already delegated can still continue somewhere else.
When you reconnect, the strange part is not that an AI wrote some code. The strange part is that work accumulated while you were not watching.
Your personal time and the computer’s working time no longer have to overlap completely.
2. “I’m not needed” is half right
The first role becoming optional is “the human who must stay glued to the screen.”
AI coding systems can take on work such as:
- investigating bugs
- changing code
- running tests
- organizing documentation
- handling independent tasks in parallel
Current OpenAI Codex workflows are explicitly designed for parallel agent work and background tasks.
So development continuing while you are doing something else is no longer a weird trick.
But the human role has not disappeared.
Someone still has to decide what should be built, what should be changed, how much authority the agent gets, and whether the result should be accepted.
The operator becomes less important.
The reviewer and decision-maker become more important.
Whether that counts as a promotion or a demotion is open to debate. It definitely involves less keyboard time.
3. The human job moves from typing to judging
The important shift is not just faster coding.
It is a change in where the work happens.
The older flow was:
think → write → research → rewrite → test
With agents, the flow becomes closer to:
define the goal → split the work → delegate → inspect the results → accept, revise, or reject
The bottleneck moves too.
Instead of “How fast can I type and implement this?” the harder question becomes “What exactly counts as correct?”
The faster the agents become, the more valuable clear judgment becomes.
4. A task on your home PC is not the same as a cloud task
This distinction matters.
A task running on your own computer can keep going while that machine remains powered, awake enough for the workflow, and reachable under the relevant settings.
Codex Cloud is different. It runs on OpenAI-managed computers. OpenAI’s current documentation says cloud tasks can continue even while your own computer is asleep.
OpenAI also describes remote access to a task running on your own computer as a separate workflow from Codex Cloud.
So these two experiences may look similar from a phone or laptop:
“My home computer kept working.” “The cloud task kept working.”
But the machinery underneath is different.
The interesting part is the same: the human does not have to remain physically beside the machine.
5. Autonomy does not mean “leave everything alone”
This is where the fantasy can become dangerous.
AI can work quickly.
It can also move quickly in the wrong direction.
If the goal is misunderstood, an agent can make a large number of perfectly energetic changes that solve the wrong problem.
Two agents can touch overlapping code and create conflicts.
A test suite can pass while the result still fails the actual product goal.
The answer is not constant human surveillance. The answer is containment.
The more autonomous the system becomes, the more important it is to make mistakes cheap and reversible.
6. Practical guardrails for autonomous development
A useful setup usually includes boundaries such as:
- keep tasks small and well scoped
- isolate independent changes
- run automated tests
- avoid giving unnecessary secrets
- separate code changes from production deployment
- keep a readable history of changes
- provide a way to stop the process
- review the diff and test results before accepting the work
This is not about “trusting AI” as a personality test.
Let it move fast where failure is cheap.
Require human confirmation where failure would be expensive.
OpenAI’s own Codex Cloud guidance also recommends reviewing the changes and test results before using the work.
7. The biggest change is not the PC. It is your time.
The real story is not “AI wrote code.”
That is already normal enough.
The deeper change is that time away from the computer no longer has to mean zero progress.
You can travel. Eat. Sleep. Think about something else.
Meanwhile, already-defined work can continue.
When you come back, the main question becomes less “What should I type next?” and more “Which result should I keep?”
That is especially powerful for solo development because one person can supervise several parallel streams of work instead of personally executing every stream in sequence.
The human is on a plane.
The PC is at home.
Development is still open for business.
Once you see that happen, “time at the desk = development time” starts to feel like an outdated equation.
8. Conclusion
“I got my connection back on a plane and discovered that development was still happening on my home PC.”
It sounds like a small moment, but it captures a much larger shift.
AI is not only increasing typing speed.
It is increasing the amount of work that can happen without a human staying attached to the keyboard.
So “I’m not needed anymore” is not quite right.
A better version is:
“The version of me that has to do every operation manually is becoming less necessary. The version of me that decides what should happen is still very necessary.”
The worker role shrinks.
The supervisor role grows.
And the supervisor might be on a plane while the home computer is still clocked in.
References (2)
- OpenAI, “Using Codex Cloud,” checked 2026-10-04
https://help.openai.com/en/articles/20001545-using-codex-cloud - OpenAI, “Codex,” checked 2026-10-04
https://openai.com/codex/
