My Abstraction Crisis: Staying Oriented While AI Keeps Moving the Goalposts

Troels Lauritz Reese Christensen — Wed, 04 Mar 2026 05:20:50 +0000

I’ve never been able to build this fast in my life.

And I’ve never felt this unstable.

The hardest part of building with AI right now isn’t writing code. It’s watching something you’re excited to build become “already handled” shortly after you start. Not because you failed, but because the level of the game changed once again.

That mental whiplash feeds almost daily into what I’ve been calling my abstraction crisis.

AI has changed the economics of building. Execution is cheaper. Tooling is denser. What used to take weeks now takes hours. So my bottleneck isn’t output anymore. It’s direction.

What’s worth building when agents can turn half your backlog into working drafts overnight? What do you bet on when the foundations keep shifting in real time?

The moment it hits

One day I’m wiring tools together and I feel ahead. My workflow is smooth, my pace is high, and it feels like I found compounding leverage.

Then the world shifts. A capability lands. An integration becomes default. Suddenly what felt like an edge yesterday feels ordinary today, and I wake up with that tight, anxious feeling of being behind.

Here’s a concrete example. One of my current side quests is building a “brain funnel,” basically a system to capture, sort, and resurface thoughts and ideas. Then I stumble across an open-source repo that’s already a high-fidelity version of what I’m building, finished enough to make my version feel unnecessary.

It’s not that I “wasted time.” It’s that I aimed at an abstraction that stopped being scarce.

And this is where I notice my own failure mode: I start scanning instead of building. I start re-planning instead of committing. Depth feels risky, because you can feel the baseline shifting under your feet.

Focus compounds

Here’s what I keep relearning the hard way: when the tools speed up, my brain tries to speed up with them.

I want to update everything. Rebuild my workflow. Start five projects because the frontier moved and I can “obviously” do more now.

That instinct feels rational. It could also be a trap.

Because what compounds isn’t novelty. It isn’t constant reorientation. It isn’t being first to duct tape the newest tool into a workflow.

What compounds is focus.

Focus turns a week of excitement into a year of progress. Focus is how you build something deeper than a demo: taste, structure, a point of view. Some people experience focus as limitation. For me, it’s essential for survival.

A framework that helps: two modes + one budget

What’s helped me more than any productivity hack is splitting my building life into two modes, and giving myself an “abstraction budget.”

Frontier mode: explore new models, new agentic workflows, new ways of wiring things together.
Production mode: execute and ship in a chosen lane with as little novelty as possible.

The budget part is simple: I have a limited capacity for “climbing layers” before my attention gets fried and I default into reaction mode. So I try to spend that budget intentionally in Frontier, and protect it in Production.

One practical rule that falls out of this: when I feel the churn starting, I go for a run before I change anything.

Questions for other builders

I’m posting this here because I’m pretty sure I’m not alone, and I want to steal better rules from smarter people.
• Do you feel this “baseline shift” effect?
• What’s your constraint for staying oriented?
• Do you time-box exploration?
• Do you keep separate “frontier” and “production” tracks?
• Or do you have a completely different mental model?

I wrote the longer version (with more detail + visuals) here: https://trolz.substack.com/p/my-abstraction-crisis-staying-sane

Will 2026 Be the Last Year I Write Code by Hand?

Troels Lauritz Reese Christensen — Sat, 17 Jan 2026 13:22:23 +0000

Experience Doesn’t Disappear. It Moves Up the Stack

I realized something peculiar this weekend.

This year might be the last year I ever write production code entirely on my own.

Fourteen months ago, I experimented with Cursor by building an app without allowing myself to write any code. It was painful, but after three weeks, I actually got there.

This weekend, I ran a similar experiment using Claude Code. This time, the app worked from my very first prompt and has rarely had any issues since. It has been an absolute blast to build like this. I have not written any code, but I have written a lot of English.

I haven’t written any code, but I’ve written a lot of English.

Times are definitely changing. This shift is affecting everyone in our space, juniors and seniors alike. It is hard to know whether my fifteen years of experience debugging browser quirks will still be useful a year from now. It is just as hard to imagine how junior developers will earn a seat at the table in the near future.

We may not be able to see exactly what is coming, but something is definitely on its way.

The hottest new programming language is English
— Andrej Karpathy (@karpathy) January 24, 2023

Almost three years ago, Andrej Karpathy said that “the hottest new programming language is English.” At the time, it felt provocative. Now it feels accurate.

Perhaps machines will solve how we compose components in React. Maybe they will decide tabs versus spaces for us once and for all. Maybe it no longer matters whether templates live next to where they are used or in some carefully curated folder structure. Maybe the machine will show us a more efficient way to express intent and to interpret it.

If my deep dive this week taught me anything, it is this. We are writing less manual code with every passing day. If you want to build, ship, and compete this year, you are going to have to leave more of the implementation in the hands of machines.

But maybe experience does not disappear. Maybe it just moves up the stack.

Not into writing more code, but into knowing what to build, how to ask for it, and when the answer looks right or subtly wrong. The keyboard matters less. Judgment matters more. Taste matters. Architecture still matters.

And if this all leads us toward Judgment Day, at least it comes with perks. No more bikeshedding. No more tabs versus spaces. And no more arguing about where to put the damn components.