The Liberation of Hard Skills
Stop clinging to what's measurable, aim higher and embrace the craft

The silence here wasn't a lack of thoughts, it was too many of them. A couple of existential crises later, here is how I think AI progress will change engineering.
I always prided myself on doing hard things others can't. Focusing longer, learning faster, working harder. I embraced AI fully, yet kept clinging to my "hard skills". Agents now understand the bug, but I'll tell them how to fix it. They write the code, but I have to unslop it. The code is clean, but I still have to review it...
Then it clicked. Humans are good at solving clearly defined problems, so a machine modelled after our brain, but with way more FLOPS, will always be more rigorous at it than we are. Competing on clearly definable (RL-able) tasks is a losing game.
The real work is deciding which problems are worth defining.
Letting go
Looking at the general workforce, hard skills like remembering syntax, writing regex, translating documents, bookkeeping, legal research or reading X-rays are all clearly definable, and therefore RL-able.
Skills were valuable because they were hard for humans, not because they were hard to define.
Making them a shitty moat, but also senior workers were never defined by them in the first place.
Senior engineers were never paid for their ability to write code. They were paid for:
- Context: knowing why the system looks the way it does and what was tried before
- Taste: recognizing a good (or bad) abstraction, design, architecture...
- Communication: turning vague wishes into clear problems, aligning stakeholders, saying no
- Sequencing: prioritizing, when everything feels urgent
Code was the output, never the value. And these skills become increasingly valuable. When building gets easier, so does wasting resources on wrong things.
Now everybody needs to act more senior. Build dependable, trustworthy systems, agents and software factories, so you can stop worrying about execution and focus on orchestration, architecture, users and problems, not on the perfect type for a parameter on line 264.
After all, nothing changed
What actually set me apart was never the hard skills, but extreme curiosity, ambition and agency.
With agentic capabilities increasing continuously, we have to recalibrate our ambition and our timelines. What used to take weeks should now ship in hours. But ambition isn't limited to time. Complexity can grow too. That doesn't mean glossing over actually learning and understanding. This step is just faster now, as the text files changed from TypeScript & Terraform to markdown & mermaid. And that's the best part. I can get my hands dirty quicker, iterate more rapidly and learn even faster than before.
Even pre-AI no topic really felt daunting but with agentic superpowers everything seems approachable, now the world truly is your oyster only limited by your creativity and ambition. But honestly this was always the case.
This amplified ambition doesn't stop at the individual. Small teams can achieve more than ever before, and a
single engineer can now oversee and operate systems that used to need a whole team. The ceilings just moved up dramatically.
Your framing shapes outcomes and mood
With agents being the great amplifiers of our time they also amplify our bad traits (laziness, impatience, carelessness just to name a few). Yes you can "slop" code, slides, messages, designs... but if you approach creation like this you set the ceiling and the feeling. If you want to create something truly great there is no substitute for bringing your taste, attention, understanding, love and passion if you sub out thinking you sub out yourself, and worst of all, the fun.
AI didn't take away the craft, it took away excuses to only be good at its most measurable parts.