Webflow's Ben Haefele: the hard part is no longer writing the code

AI just made the building part cheap. That did not make your job easier. It moved the hard part. The question is no longer "can we ship this?" It is "should we, and is it any good?" That shift is already showing up in job titles, pay bands, and promotion ladders. Let me catch you up.

The deep cut

• Cheap execution makes judgment the scarce thing. Webflow's Ben Haefele put it plainly: the hard part is no longer writing code.
• Taste does not come from a model. Gamma's Operation Falcon kept humans watching customers live, automating everything except the insight.
• A job is tasks, not a title. Anthropic found AI handles routine work first, which guts the junior rungs where people learn.

What you are actually paying for now

Across 12,500 PM job descriptions, the top skill was not RAG or agents or prompt engineering. Ankit Shukla's read is that it was judgment: knowing which problems deserve AI at all. More than 30% of PM roles now ask for AI skills, and those roles pay 15 to 20% more in the US.

The same point came from the stage. At ProductCon, Jeff Chow of Miro said building the right thing is becoming the bottleneck. Many teams solved parts of execution. More output, more noise, less clarity about what any of it is for.

So when you write a req or a promotion case, stop leading with throughput. Lead with the call. Can this person pick the problem worth solving and recognize good work when it lands?

The trap of cheap prototypes

When building was expensive, bad ideas died because nobody could justify the cost. That friction did real work. It forced you to prioritize.

Now a prototype takes an afternoon. A bad idea does not die. It becomes a prototype, then another, then another. Suddenly you can generate more ideas than you can evaluate. Prioritization gets more important, not less, because you have more things to decide between.

Jeetu Patel of Cisco closed that day with the sharpest version: in a world where you can do almost anything, what you choose not to do becomes the point. That is a review-room discipline, not a tooling problem. Bring a kill list to your next roadmap review, not just a build list.

Taste comes from proximity, not a dashboard

Here is the catch on all this judgment talk. Judgment has to come from somewhere, and it does not appear because you bought a better model.

Diana Kimball Berlin of Gamma ran Operation Falcon: dozens of employees joined over a hundred customer calls in weeks. AI handled scheduling, transcription, first-pass synthesis. Humans watched customers use the product live. Her rule was "automate everything except the insight." They caught voice and microphone problems that aggregate data would have hidden.

The risk is you get very good at processing summaries of reality while drifting away from reality itself. Protect the hours your team spends watching real users. That time is not overhead now. It is the input that makes every downstream call better.

The rung that is disappearing

The uncomfortable part sits under the optimism. AI eats the routine tasks first, and those are exactly the tasks juniors used to learn on.

A Harvard study cited by Towards AI found junior employment fell about 8% at AI-adopting firms over six quarters, while senior headcount held or grew. Anthropic recorded a roughly 14% drop in hiring for 22-to-25-year-olds in high-exposure jobs. If nobody hires for the simple work, where does the next generation's judgment get built?

This connects to how you promote. One design ladder treats promotion as recognition that someone already operates at the next level, with behavior as the evidence. That logic still holds. But if AI has stripped out the tasks where junior people prove themselves, your ladder needs new rungs that reward judgment and customer proximity early, not just reliable delivery.

Three questions for your team

• When you read your last three open reqs, do the top bullets describe judgment and problem selection, or just output and tools? If it is output, rewrite them.
• How many hours this month did your team spend watching real customers use the product live, with no AI summary in between? Put a number on it.
• If AI is eating the tasks your juniors learn on, what is the new first rung on your ladder, and who owns rebuilding it before your next promotion cycle?