Last year I assumed I wouldn't have much new to say about programming twelve months later. A fresh framework would drop, a language would be declared dead, and the discourse would move on. None of it felt paradigm-shifting.
What actually shifted is the layer between intent and outcome. LLMs got good enough that typing the code — the act of translating what you want into syntax a machine will accept — is no longer the expensive part. That changes the job. Just not in the way your timeline would have you believe.
You were never a coder
"Coder" never quite held up as a title. Code was always a language we learned so computers could act on our intent, and the history of software engineering is a slow collapse of that translation layer. COBOL, SQL, Python, Ruby — every generation moved closer to plain English. LLMs took one more step in the same direction. A bigger step than most of the ones before it.
The point was never the lines. It was the outcome. We just happened to need a human at a keyboard to get there.
That part is thinning out. Coding as a job is going away. Software engineering isn't. Fewer keystrokes, more judgement. Knowing what to build, for whom, with which guarantees, when to push back.
Beware the recycled jargon
If you've been on LinkedIn, you've seen a wave of tech vocabulary dressed up to sound new. Most of it isn't. A short field guide to the ones worth discounting:
Spec-driven development. Real concept. Also not new — it's the grandchild of model-driven development, which Martin Fowler wrote about twenty years ago. Anyone who's shipped Agile in a mid-sized company has been working from acceptance criteria for a decade. The real shift is that the machine can now do the execution while the human does the spec and the review. That's the delta. The branding isn't.
Disposable code. People frame this like AI made code disposable. Code has always been disposable. Most of what any of us wrote in 2018 has been rewritten, replaced, or deleted. AI just collapses the feedback loop and forces the disposability into the open. Stop tying your identity to the lines.
Product-minded engineers. Engineers being involved from ideation to launch is how most healthy product teams have always worked. The rebrand into "full-cycle developer" is marketing, not a new discipline. And asking one person to absorb design, research, PM, and analytics alongside the engineering is a burnout pattern, regardless of how it's framed.
The repeating pattern: real shifts are happening, but the naming committee keeps declaring victory on changes that were already in motion.
Every team now has an AI denier
2026 will be rough on engineering teams. LLMs are past the novelty phase and well into "this is how the work happens now." That means AI adoption stops being an R&D experiment and becomes a line item in the quarterly plan.
Expect at least one engineer on every team to push back. The reasons vary — fear of obsolescence, worry about skill atrophy, a purist aversion to committing code they didn't type, sometimes just stubbornness. Some reasons are defensible. All of them are corrosive when leadership doesn't take a position.
"You can use AI if you want" isn't a directive. It's a shrug. Teams that adopted Agile well had clear ceremonies, definitions of done, review gates. Teams adopting AI well need the equivalent — which tools, for which tasks, with which verification step before merge. Without that, the denier becomes the loudest voice and the team drags.
Timelines shrink, and autonomy spreads
Two related shifts:
PMs and designers get more of the pipeline. A designer with a working design system can submit a frontend PR and hand it to the engineer for polish. A PM with a spec tool can turn a loose idea into an estimable story without blocking on a standup. That rebalances who owns which step.
Two-week sprints age fast. When idea-to-PR shrinks, the ceremony around idea-to-PR starts to feel like overhead. My bet is Kanban becomes the default for most teams by mid-year. Pick a task, do it, ship. No retro on why you missed the sprint goal, because there isn't a sprint.
The corollary: inflating estimates to carve out breathing room gets harder. You either get faster or you look slower, relative to a team that adopted the tools.
You don't do less — you try more in parallel
The most misread piece of AI at work is the "automation saves you time" framing. That isn't what's happening. What's happening is that experiments that used to be expensive are now cheap, so you run more of them.
Prioritisation becomes less of a high-stakes committee meeting. You don't agonise over which two features to try this quarter — you try eight and let the data pick. Teams that frame AI as "sip coffee while the machine works" are missing the point. The machine works and so do you. Just on ten things instead of two.
Tests quietly become cool again
For years I've watched teams undersell automated testing. Writing tests was the chore nobody volunteered for. A failing test was treated as a blocker, not a safety net.
That story flips when most of the code in a PR was written by a model. Tests stop being overhead — they become the only remaining human-in-the-loop. Not because humans write the tests (AI writes plenty of those too), but because a human still has to confirm the test is checking the thing that matters. That's the new contract: we trust the generation if we trust the verification.
If you've been letting your test pyramid rot, 2026 is the year to fix it. The cost of regressions in an AI-authored codebase is higher, not lower.
The identity crisis is the real thing
Everything above is mechanics. This part is the harder question.
When you've spent years building an identity around writing code, and the typing part compresses to maybe 10% of the day, what's left? Some engineers will coast past it without ever confronting the shift. Most will hit a wall. Leadership that doesn't plan for the wall will lose people it didn't need to lose.
The frame I've settled on: we were never just coders. The editor was always the tool. The thing we were actually doing — the thing that still needs doing — was solving real problems for people who couldn't solve them alone. Nothing about 2026 changes that job. It just changes the shape of the day.
Carry that with you.