Make haste slowly
Jan 29, 2026
Globally, by early 2025, 29% of all newly written software functions relied on AI assistance“AI-assisted coding reaches mainstream adoption.” Science, 2025, www.science.org/doi/10.1126/science.ado5344. (up from 5% in 2022). The unspoken implication: fire those pesky programmers and replace them with an API call. AI acolytes believe that all software development, indeed all knowledge work, is about to become obsolete; skeptics are convinced that over a trillion dollars have been wasted and a dot-com style collapse is imminent. All roads, it seems, lead to mass unemployment and a takeover by skynet.
Never a confident bunch, senior engineers are worried about a rapid devaluation of skills that were acquired over a lifetime. Worryingly, enrollment in undergraduate and graduate courses has decreased. If one can “vibe-code” software, then it’s just a matter of wishing the AI genie to solve problems. Learning is a fool’s errand. So do we look forward to a future with greater unemployment, fewer apprentices to the art of software development, and rapid loss in institutional understanding of complex human spaces? All cheered on by ungodly sums of VC money. Conspicuous by its absence: meaningful conversation about societal good!
The roughly dozen folks at Answer.AI, led by Jeremy Howard and Eric Ries, are making a contrarian bet: what is needed most in this moment is a return to curiosity; a return to craftsmanship. They believe AI affords the ability to cut deeply through layers of complexity. There is an opportunity for humans to deeply comprehend machine-scale systems. Their answer: a generalized compute platform called SolveIt that integrates AI with literate programming.
Roman coins minted in the time of Emperor Augustus bore the emblem of the dolphin and anchor, symbolizing “festina lente” — make haste slowly. You may know the modern adage: “slow is smooth and smooth is fast.” SolveIt is a platform that enables iteratively solving problems while pair programming with an LLM. One step at a time. Named after a problem-solving method from the mathematician George PólyaPólya, George. How to Solve It. Princeton University Press, 1945., SolveIt feels like an evolution of Jupyter notebooks (which allow exploration by combining code, data, and notes into a living document). In a marked departure from the “vibe-coding” tools, SolveIt (by design) slows down thinking and encourages exploration as the software is built.
By pure serendipity, the enrollment deadline for the second SolveIt cohort was about a week after I quit my job (my first break in about 25 years!). Having missed Jeremy Howard’s previous courses, signing up was a no-brainer. There is something immediately different about this course. A few hundred folks from various timezones are on. The sessions are unpolished; perhaps the lack of gloss is the point. But there’s shared curiosity, almost like we wait for the results of a chemistry experiment. There is attention to detail. And there is kindness. Howard and Johno Whitaker run through problems from Advent of Code. A longtime skeptic of LeetCode-style challenges, I had written these off. But suddenly the puzzles are fun. Taking a problem, breaking it down into the smallest elements, and building it back. I am more traveller, less a tourist.
These sessions traverse large distances. In sessions lasting roughly an hour (sometimes too much fun to mind the clock) we deeply explore how agents like Claude work. Building them from scratch. We set up cheap infrastructure outside the popular hyperscalers. We dip into linear algebra. I was never good at this, but suddenly I’m building intuition about matrix math. Eric jumps on and talks about harsh realities of the business side that plague innovators. Skills I previously assumed were the domain of polymaths and giants feel accessible. It has been years since I had written code. But like most managers, I did not start my career because of a deep love of managing managers of software engineers. Yes, this is work, but it is fun to write code again.
Anthropic CEO Dario Amodei says, “Humanity is about to be handed almost unimaginable power, and it is deeply unclear whether our social, political, and technological systems possess the maturity to wield it.”Amodei, Dario. “Machines of Loving Grace.” 2024, darioamodei.com/machines-of-loving-grace. A few minutes on LinkedIn will tell you that you are obsolete and your peers have already solved all meaningful problems. AI skeptics will call LLMs a farce. Perhaps Amodei is right. Regardless, neither side offers a path for the motivated and curious (and likely terrified). As a profession, coders experience high levels of imposter syndrome. The antidote isn’t more AI - it’s more human agency. When engineers approach problems with genuine curiosity rather than a rush to ship, they build systems they actually understand. Craftsmanship means taking small, deliberate steps. It requires breaking every hard problem into pieces small enough to hold in your head. This produces maintainable code because you can’t maintain what you never understood. It rebuilds confidence because imposter syndrome thrives on mystery, and curiosity dissolves mystery. The engineer who asks “why does this work?” doesn’t need mystical incantations.
Reject the noise. There are real problems waiting in the world.