Ladders
Chaos is a ladder
The last time I thought seriously about engineering career ladders was at Mixpanel. Writing rubrics, sitting through calibrations, debating what separates a strong mid-level from a staff engineer. At Pocus we skipped public levels entirely. So it’s been about three years since I’ve had to sit with a rubric and ask whether it actually measures anything real.
My thinking around levels evolved quite a bit at Mixpanel and coming back to it, what strikes me is how much of what I used to believe about ladders no longer holds. Most ladders haven't caught up to the changes in the last year and companies are running promotions against rubrics that are quietly measuring the wrong thing.
Judgment & Ability
Ladders do a lot of work inside a company. They're comp scaffolding, calibration mechanisms, recruiting signals for candidates trying to figure out where they'd land, and legal cover for pay decisions. All of that is real. But the thing that justified the whole structure was filtration.
Most large-company ladders gave you a finite window to reach the terminal level. You could clear a level early if you demonstrated the ability. Miss the window at the junior levels and you were managed out. Companies were incentivized to invest in engineers that grew quickly, and engineers were incentivized to always keep pushing for the next level because compensation, influence, etc. were all tied to levels.
If you peel back the layers of any rubric, every ladder was testing for two things.
Ability, meaning what an engineer can do.
Judgment, meaning whether they know what good looks like.
Ladders didn't talk about these axes directly. They talked about scope and impact and technical contribution, the visible outputs. But the outputs were proxies, and the reason the proxies worked was that the two axes were correlated in practice. An engineer who could write good code could usually review code well. An engineer who could build a good system could usually tell when someone else's system was bad. Ability was the visible thing, judgment was the invisible thing, and filtering on the visible one was a reasonable proxy for the invisible one. The ladder worked because the correlation held.
Senior is the new entry level
AI has broken the correlation. Ability is now available to anyone paying $200 a month. Judgment is still somewhat scarce. You can find engineers with high ability and no judgment, which shouldn’t have been possible in the old world and is now routine.
The old senior bar is now what companies expect from anyone they hire. Every rubric written before AI assumed that a senior engineer had absorbed years of judgment on top of strong technical skills. That’s no longer how it works. A capable junior engineer can ship complex features with Claude’s help, pass code review, and look senior on paper, while missing the judgment to see how the change lands in the rest of the system.
The visible signal the ladder used to filter on no longer carries the information it used to. Filtering on ability now selects for engineers who can direct models, not for engineers who know whether the model is right.
The path to developing that judgment has compressed too. The feedback loops that used to take days now take hours. A junior engineer used to spend days writing code, testing it, running into edge cases, and a senior engineer would respond with fifteen minutes of review. That whole arc now happens in a single sitting, and capable models can give code review feedback at something approaching a senior engineer’s level. The engineer who was going to develop judgment over five years can develop comparable judgment in a year or two, if they engage with the cycles instead of just shipping through them. The bottleneck isn’t time anymore. It’s whether the engineer is paying attention to the right things. Even in cases where you still need calendar time to watch decisions age, you’re accumulating more decisions per unit of calendar time for those lessons to attach to.
Compression & Expansion
Ladders need to compress and expand at the same time.
Start with compression. The old ladder had many levels because each one took years. You needed fine-grained checkpoints to track progression through a multi-year window and to anchor comp along the way. When the old senior bar becomes the baseline for anyone you hire, senior can't be terminal. And if the learning loops have compressed, there's no reason for a junior engineer to take four or five years to reach what used to be senior. Most ladders have historically responded to "engineers stuck too long at a level" by adding levels. Uber introduced L5B because people sat at L5 forever. A lot of companies added a Senior Staff rung between Staff and Principal for similar reasons. That pattern has to invert. The fix now is fewer levels, not more, because the time at each level is shrinking.
The expansion is about what counts at each level. T-shaped engineers, meaning depth in one area and working knowledge of others, were the baseline everyone aspired to. That's not enough anymore. The engineers who matter in the new world have depth in multiple areas and genuine fluency in adjacent functions like product, sales, design, etc. Multi-pronged T-shaped. This also changes what counts at each rung. A lot of ladders reserve cross-org impact for Staff and above. In the new world, that bar applies far earlier. A mid-level engineer who can't operate across product and go-to-market boundaries is missing something that used to be optional and is now baseline.
A good ladder in the new world has fewer levels, sharper boundaries between them, and expects more at every level than any ladder does today.
Fin
Ladders aren’t going to update themselves. No one inside the system is incentivized to change it. The people writing and maintaining these ladders are usually the same people who benefited from them, and asking them to rewrite the rubric is asking them to relitigate their own careers.
Most companies will keep the existing structure and quietly absorb the cost of measuring engineers against a world that no longer exists. The orgs that rewrite the levels and raise the expectations now will have the engineering teams everyone else is trying to copy in a few years.