AI has changed what my job involves. More of it now is specifying the work, directing the tools and checking what comes back.
In January 2026 I refactored a Laravel codebase, with 78 Actions and more than 3,000 tests, using an AI agent running in a loop overnight. Done by hand, the migration would have taken weeks of tedious, error-prone work.
It also failed in ways I didn’t expect: it invented categories I hadn’t asked for, and reported tasks finished that it hadn’t touched. I wrote about the gotchas in detail. The AI did the work in the end, but only once I was better at directing it than I was at doing the work myself.
The debate
You’ve seen the takes. “AI will replace developers by 2025.” “AI is just autocomplete with better marketing.” “Learn AI or become obsolete.” “AI can’t do real work.”
Both camps get it wrong. One side assumes that what AI can do makes people irrelevant. The other assumes today’s limits are permanent. Each side is too busy defending its position to look at what’s happening.
AI is a tool, and it’s changing what work looks like. People who learn to use it well will outpace those who don’t.
How I use it
I use AI every day, as part of how the work gets done. The skill is knowing which model, which prompt structure and which checks suit each problem.
On a refactor, the work has moved from “write this function” to “specify exactly what this function needs to do, how to check it works, and what failure looks like”. That’s harder than writing the function, and faster once you’re good at it.
That makes prompt engineering part of the job, because the quality of the direction sets the quality of the output.
AI widens the gap between good and mediocre work. A developer who understands architecture, testing and system design can direct it to produce good work quickly. A developer who doesn’t will produce the same mediocre work faster, or broken work they can’t check.
The competition
For most businesses, the risk is a competitor who worked this out six months earlier. That might be a supplier who delivers in two weeks what used to take six. Or it might be a business that improves its product faster, because its developer automates what others still do by hand.
What adapting looks like
Adapting means specific changes in how you work.
Knowing each model’s strengths. Claude reasons well but can be wordy. OpenAI’s GPT models handle some structured outputs well. Smaller models are fast and cheap for simple tasks. Knowing which to use, and when to switch partway through, gets more useful with practice.
Building checks into everything. AI sounds confident even when it’s wrong. The Ralph loop showed me that “tests must pass” invites an agent to find the cheapest path to green. Say how as well as what, and check at every step.
Treating prompts as code. Version them, test them and review them. A prompt that works today might drift tomorrow.
Building it into the workflow. The gains come from using AI in how you research, draft, refactor, review and ship, more than from asking ChatGPT the odd question.
Staying accountable for the output. AI generated it, but your name’s on it. If you can’t check the work, you can’t ship it, so you need to understand what the AI produced well enough to catch its mistakes.
For business owners
You may be wondering whether AI is worth anything to your business, or just noise.
One example: testing an idea used to be expensive. A concept needed mock-ups, perhaps a prototype or a demo environment, all before you knew whether the direction was right. AI can now produce visuals, draft interfaces and build working prototypes, and revise them on feedback in hours rather than weeks. You can explore three directions before putting budget into one.
The ideas that survive early exploration get proper investment, and the ones that don’t are dropped cheaply. Designers and developers still build what goes ahead.
What to look for in partners and hires:
- They can explain specifically how they use AI in their work.
- They’re faster than they were two years ago, and can show why.
- They talk about checking and quality as well as speed.
- They’re honest about what AI can’t do.
Red flags:
- “AI will handle it” with no explanation of oversight.
- Speed promises with no mention of how quality is maintained.
- Dismissing AI entirely (“we don’t need that”).
- Treating AI only as a way to cut costs.
The value is more work at the same quality, or better work at the same speed. Be wary of anyone selling AI only as a way to pay less for the same output.
For developers
If you’re anxious, that’s reasonable. These are the things that help.
Start with your existing workflow. Where do you lose time: boilerplate, research, documentation, code review? Pick one, find an AI tool that helps with it, and get good at that before you expand.
Learn prompt engineering properly. Structured prompts, few-shot examples, chain-of-thought reasoning and output formats carry over to every model and every task.
Build checking habits now. Developers who do well with AI are quick to check its work. Learn to write prompts that produce output you can check, and to spot when AI is confidently wrong.
Stay technical. A developer who understands database indexing can direct AI to write fast queries. One who doesn’t will ship slow queries faster. Architecture, testing and system design matter more now.
Ship something with AI this week. Use it on real work rather than a toy project, make a mistake and learn from it.
Where that leaves me
Version control, automated testing, cloud infrastructure and CI/CD each changed the job. At each shift, people worried about becoming obsolete, and the ones who learned the new tools became more valuable.
AI follows the same pattern, faster: learn the tool, direct it well, and stay accountable for what it produces.
I’m not worried about AI taking my job. I use it to do the job better.
AI is part of how I work on the Laravel, WordPress and Shopify projects I take on.
Published , updated .