Stages of grief in AI Development

Recently I read a tweet from the creator of Node.js that roughly said: "the era of humans writing code is over." It hit me harder than I expected. It sent me straight back to 2009, when Ryan Dahl announced Node at JSConf. That moment felt like a beginning; and we built a company on it. This moment felt like an ending.

The full tweet:

This has been said a thousand times before, but allow me to add my own voice: the era of humans writing code is over. Disturbing for those of us who identify as SWEs, but no less true. That's not to say SWEs don't have work to do, but writing syntax directly is not it.

— Ryan Dahl (@rough__sea) January 19, 2026

Realtalk: I think this tweet says two things that are 80% true now, and will get to 99% before too long.

As someone who spent most of my life getting good at coding, I went through the five stages of grief on this. Sadness at a life investment getting depreciated in the marketplace, and in my self-esteem. Fear of solo competitors eating our lunch. Anxiety about a weird world where nothing is real, job displacement, increase in the gap between rich and poor, social unrest.

It's natural to have a negative response. I'm seeing others go through similar things — denial, anger, bargaining, depression. For your sake: don't get stuck in one stage. Reality is changing and we need to adapt. Others are getting leverage from these changes. We can't, so long as we're combative or stuck in denial.

I thought AI would take the fun away, and honestly it did for a while. Waiting, fighting slop, waiting more. Still a net productivity benefit but not enjoyable work.

But in the last few weeks, the tools and workflows, matured past that threshold. Less waiting. Less slop. And I found that same lego-like creativity again — now at a higher level of abstraction than if statements and for loops.

Here's what I think about now:

• Context design: An LLM is like a big alien brain that spent its life in a prison cell reading every book. What windows do we poke in the cell? What tools does it need?
• Sequencing: I can build all these components — in what order do I arrange them for maximum leverage, each amplifying the next?
• Guardrails: How do I ensure good outcomes instead of spending 1 day building and 2 days cleaning up? Can I give the agent 'hard things' to bounce off of? Which tests do we need first?
• Iteration speed: What can be sped up to cut cycle time dramatically?
• Human checkpoints: At which strategic points do I still interject, handroll, or vet?
• Formalization: What can become a procedure doc/skill to avoid friction next time?
• Division of work: What will be the least regretful division of work longer term? This is true as much about components as division of work between agents from different houses (codex for backend, claude for front, gemini for visuals), and what tools they get to use (a browser to check their own work).
• Scaling: Where is the true bottleneck — worktrees, machines, imagination, CI time, sanity? How do I avoid an agent just always adds more code, resulting in a colossal maintenance hell down the line. How do I effectively force it to re-use what we have? How does it know about this without polluting context?

These are still rewarding puzzles. Different puzzles. But seeing it work out and breathing life into larger systems is just as, and sometimes maybe more rewarding on the multi component/project vs multi line/function level.

But we won't get there by categorically refusing. It may require sitting down, taking a deep breath, accepting the discomfort and the new reality, before we can start getting leverage from it.

Speaking as an employer: there is a place for SWEs who can herd agents and vouch for outcomes, for a long time to come. The leverage I can get is wild, but there's a real limit to how many agents I can safely run in parallel. Engineers who can think about these puzzles and take ownership of outcomes are absolutely worth their weight in gold to me — and likely to other employers too.