A nurse who knows exactly what confuses families at discharge can now build the tool that fixes it. They don’t need to hire a developer or learn to code, because an AI coding agent does the typing. What’s left is knowing what those families need, and the nurse knew that years ago. A teacher who has watched the same lesson stumble for years is in the same position.
I’ve built WordPress plugins for over a decade. I think the world gets much better when people who feel deeply and think clearly can ship their own ideas, and agents now let far more of them do it.
Who Gets to Build Now
Shipping software used to take a team or a lot of spare hours. My guess is that plenty of good ideas from observant people stayed in notebooks because of that.
The person I have in mind feels a lot and thinks well at the same time. The feeling side notices that a form confuses an older customer, or that a dashboard makes a nervous person more nervous. The thinking side turns that observation into a design that holds together.
That pairing used to stall at the build step. Someone could see exactly what should exist and explain it clearly, then wait months for engineering time. Or they handed it to a team that saw it differently.
With an agent doing the typing, the person who had the idea can take it all the way to release. How fast you could produce code used to filter out a lot of depth and sharp thinking. That filter is mostly gone.
Many of these people have demanding jobs or young children at home. Their sense of what would help others was there all along. Agents give them back enough hours to act on it.
I live in Cascais, Portugal, with my family. A parent who knows how a school morning falls apart will design a better tool for it than a team working from a market report, and that parent can now build it.
The same goes for a shop owner who hears the same customer question every day. They understand that problem better than any outside team, and they no longer need one to build the fix.
The Job Moved
If you build software, your daily work now looks a lot like leading a team whose members are models. The job is helping the agents do better, and improving the setup each time they fall short.
I pay for five AI coding agents: Claude Code, Codex, Grok Build, Google’s Antigravity CLI and Cline. They all load the same skills, rules, MCP servers and memory about my work. Whichever one I open starts from the same context.
That shared layer is where the work sits. When an agent misses something, the fix that lasts is the missing instruction, skill or check, written into the shared layer. Once it’s there, every agent picks it up on every later run.
Good team leads handle new hires the same way. When someone misses something, they ask what the system failed to tell that person and change the system. With agents the loop is faster, and one written lesson reaches all five at once.
The skill involved is close to briefing a capable colleague who starts each session knowing only what you wrote down for them.
The models underneath will keep changing. The skills and rules carry over to whichever ones come next, so I think that layer deserves more care than any single line of code.
Briefing an agent well sharpens your own judgment, because you have to know what you want before you can ask for it. A lot of “I’ll know it when I see it” is really a gap in your own thinking. Writing the brief closes it, and your taste gets clearer as you go.
Patience and attention to detail still pay. The sharpest instruction usually names the small thing everyone else skipped.
Everything Around the Code Sets the Pace
Writing the code used to be the slowest step in building software. Agents now write it fast enough that the pace comes from the steps around it.
A spec has to say what correct looks like. CI has to catch what the agent got wrong, and then a person reviews the result before it ships. Those steps always existed, hidden inside the long stretch of typing.
Adding an agent to the same tickets and pull requests helps a little, and then the old process becomes the ceiling. The bigger gain comes from rebuilding the workflow around the agents. Specs get precise enough that a model can check its own work against them, and tests become how you trust the output.
Review changes too. When the checks before it have done their job, a reviewer can skip the easy parts and spend attention on the hard ones.
Spec writing is where care shows up first. To say what correct looks like, you have to decide what the customer needs, and you can’t do that without understanding them. A model can’t ask the customer what they meant, so someone has to have done that before the spec gets written.
Writing it down often shows you the idea wasn’t finished yet. So I think this shift favors the careful thinker who also cares about the person on the other end. That’s the person who writes the definition of done well.
Taste Decides What Deserves to Exist
Ask an agent for ten versions of a landing page and you’ll have them in minutes. Choosing which one should ship is a human call. A model will produce the tenth variation without complaint, while a person with taste may decide the second was better.
A person with taste can also explain why. That explanation becomes a better brief for the next round.
When a mediocre product costs almost nothing to build, there will be many more mediocre products. That raises the value of judgment. I think the builders who stand out will be the ones willing to throw a draft away and make it again, even when it technically works.
Taste grows from looking closely at things and noticing what bothers you. It also grows from using a product the way a customer would. Someone who has felt confusion and relief can predict how a stranger will feel using what they made.
These are the qualities I value most, and they’re what I hire for:
- Taste, and the care for people that informs it
- Passion for what you’re building
- Curiosity about how things work and why people use them
- A deep drive to improve, and to do things differently and better
I’d rather work with someone who has those four and little technical training than with a strong engineer who has none of them. In my view tools are easier to learn than care for a stranger’s experience when nobody would notice the shortcut.
Picture two people given the same agents on the same morning. One accepts the first output that runs. The other reads it as a customer would, catches the clumsy sentence and sends it back with a sharper brief. Give it a year and I’d expect their products to look nothing alike.
People Stay in Front of Customers
Agents can take the repetitive build work, like the fortieth version of a settings page. That frees people to stay in conversation with customers who have real problems.
On a customer call you hear the hesitation before a sentence ends. You notice what they’ve stopped asking for. That’s how you learn what to build next, and it needs a person on the call.
Builders used to vanish into the code for weeks and come back with something half-stale. Now they can stay in the conversation while the agents work, then turn what they heard into a clear brief. The brief is better for it, because it carries what the customer actually said.
That changes what a small team can be. A few people with strong judgment and well-taught agents can serve a large market and still answer every customer personally.
Companies tend to lose that kind of warmth as they grow. A small team with good agents can keep it much longer, and I think customers will notice and reward the builders who do.
When building gets cheap, more people build, and more of what gets made comes from people who feel deeply and think clearly. They’ll build tools shaped by someone who saw the problem up close. I hire for those traits.
If you’ve had an idea for years and never had the hours to build it, try building it now with an agent.
Related
Leave a Reply