If you have worked on both sides of this — inside a Bay Area team and inside a distributed one — the rate difference is hard to explain honestly for a while. The distributed engineers were not worse. Often they were better on pure implementation, because they had fewer meetings.
So what was the four-times rate difference buying?
It took me a few years to be able to answer that in a way I actually believed, and the answer turns out to be narrow, specific, and much more defensible than "brand."
The scarce thing is calibrated pessimism
Not intelligence. Not skill. Calibration.
Here is what it looks like in practice. Someone proposes an approach in a design review. It is reasonable. It works. And a senior engineer in the room says: this is fine until roughly here, and at that point it becomes very expensive to change, and here is the specific way it becomes expensive.
Then — and this is the part that matters — they are right. Not right in a general cautionary sense. Right about the specific failure, in roughly the timeframe they estimated.
That accuracy comes from having been present when it happened before. Three times, at three companies, in slightly different forms. It is exposure, not intellect, which is why it correlates with place: a dense ecosystem of funded companies, acquirers, and people who have worked at four of them produces engineers who have seen more distinct failures per year of career than engineers anywhere else.
You cannot hire for it with a job description. You can only be somewhere it accumulated.
The thing that is not scarce
Everything else, honestly.
Implementing a well-specified component. Integrating a documented API. Building the admin surface. Writing the mobile client against an existing backend. Refactoring a module whose target shape has already been decided.
This is skilled work and it needs to be done well, and there is no correlation between doing it well and being within forty miles of San Francisco. The rate is the same though, because California compensation is set by competition with well-funded product companies rather than by the nature of the task.
That is the arbitrage most people eventually notice, and it is why almost every California engineering organisation of any size runs a hybrid structure internally, whatever their careers page implies.
Then AI arrived and made this impossible to ignore
The category that got compressed is exactly the category that was never scarce.
Scaffolding, data layers, API clients, migrations, test fixtures, mechanical refactors — the implementation volume that a wide junior base used to absorb now completes in a fraction of the time. Genuinely large, not incremental.
The category that did not get compressed is exactly the one the premium was pricing. Calibrated pessimism does not come out of a model. Neither does deciding what to build, nor the specific skepticism required to catch generated code that is syntactically clean and quietly wrong.
If anything, review got harder. More code arrives per unit time, and it lacks the human authorship signals reviewers were unconsciously calibrated on — the rushed variable name, the awkward abstraction that suggests the author was confused. Generated code is uniformly confident-looking, including when it is incorrect.
So: the cheap input got cheaper, the expensive input stayed expensive and got slightly more valuable, and the bottleneck moved from producing code to examining it.
What that implies about team shape
The pyramid — one architect, two seniors, six or seven mid and junior engineers — was an economic answer to implementation being the constraint. It is not the constraint any more.
A senior-weighted team of five will usually outperform a pyramid-shaped team of ten now, and the mechanism is straightforward: review capacity. In the pyramid, all output flows to three people who are also doing architecture and stakeholder work. The review queue develops depth, depth creates pressure to approve rather than examine, and quality erodes in a way that is invisible on a velocity chart until an incident makes it visible.
If you are on the buying side of this, the relevant consequence is that a twelve-person proposal for work a strong six-person team could do is a 2022 cost model at 2026 prices. Ask for a re-scope. And negotiate the seniority mix rather than the blended rate — blended-rate negotiation pushes vendors directly toward the junior-heavy shape that performs worst under the current constraint.
What this means if you are the engineer
Two things, and I think both are underrated.
Reviewing is now the higher-leverage skill. Not writing. The scarce activity in most teams is examining a large volume of plausible code and finding the parts that are wrong. If you want to become more valuable, become someone whose review catches things, and be explicit about what you caught and why.
Calibration is acquired by exposure to consequences. The engineers with the judgment being paid for are the ones who stayed long enough to see their own decisions play out. Two years is enough to ship something. Four is roughly where you start finding out whether you were right. Optimising a career purely for the former yields a lot of shipping experience and comparatively little calibration.
The practical version
Buy — or become — judgment in small quantity. Buy execution in volume.
That is not a cynical framing of how outsourcing works. It is a description of where the scarcity actually is, and it happens to be what well-run engineering organisations have always done, whether or not they described it that way.
The full buyer-side guide — market segments, regional differences, current rate bands, CCPA engineering obligations, California contract specifics, and twelve-month cost ranges — is here: Software Companies in California, USA.
Frequently Asked Questions
Is the Bay Area premium justified?
For a narrow set of work, yes — architecture, product direction, and technical decisions whose consequences arrive at scale. For well-specified implementation, integrations, and maintenance, there is no meaningful quality difference from a strong team elsewhere at a fraction of the rate.
What is "calibrated pessimism" concretely?
The ability to predict specifically how and roughly when an approach will fail, and to be right. It comes from having been present for that failure before, several times, which is why it correlates with ecosystem density rather than with individual skill.
Did AI reduce the value of senior engineers?
The opposite. It compressed the cost of the work seniors were least needed for and left untouched the work only they do. It also made review harder, because generated code lacks the authorship signals reviewers had unconsciously learned to read.
Why does a smaller senior team often outperform a larger mixed one?
Because review capacity is now the bottleneck. In a pyramid-shaped team, all output funnels to a few people who are also doing architecture, producing a deep review queue and pressure to approve rather than examine.
What should engineers focus on developing now?
Review skill and calibration. Being able to examine a large volume of plausible code and find what is wrong is the scarce activity in most teams, and calibration comes from staying long enough to see your own decisions play out.
Does this mean junior roles are disappearing?
The economic argument for hiring juniors as throughput has weakened; the argument for hiring them as future seniors has not. That implies fewer junior roles with substantially more mentorship attached, focused on learning to evaluate generated code rather than competing with it on volume.