Slide 01
You built something that made software fun again. Engineers who had been at this for twenty years forgot what flow felt like until your tools handed it back. That is real. That matters. Now stop making us think about which model we are on.
Slide 02
Engineer and agent are in sync. Architecture is taking shape faster than anyone can type. The tool disappears. It is just the engineer and the system they are designing.
Too many premium tokens. The agent gets dumber mid-refactor. The context window shrinks. The conversation that felt like a partner now feels like onboarding someone new mid-sprint.
Forty minutes trying to get back to where they were. Sometimes they do not. The flow state is not paused. It is gone. That is engineering output you destroyed.
The thing that felt like a conversation five minutes ago now feels like onboarding a new team member mid-sprint who does not have the context.
What model downgrade actually feels like
Slide 03
Engineers should not have to decide whether a task is "worth" the frontier model or whether they should save credits for later. They became engineers to build things, not to manage a token budget.
And they really do not want "smart routing" that silently swaps the agent out mid-refactor for a cheaper model and hopes they will not notice. They always notice. The quality drops. The context gets foggy. The suggestions go from sharp to generic.
If you can build routing that is genuinely invisible -- frontier reasoning when the problem is hard, fast-model speed when it is trivial, and the engineer never knows or cares -- ship it.
But the moment the engineer feels the swap, the moment the agent loses the thread of what they were building, they would rather pay for the frontier on everything than deal with a system that drops the ball one time in ten.
Slide 04
Engineers start self-rationing. They reach for the cheap model by reflex, even when the hard problem in front of them needs the frontier. You are training your power users to underutilize the tool they are paying for.
Engineers feel guilty about using the good model. They start treating high-quality output as an indulgence instead of the baseline. This is the opposite of what your product is supposed to do.
The engineers who will put up with this the least are the ones who churn the fastest. Your best users. Your advocates. The ones who would have sold your tool to their next company. Gone.
Slide 05
Slide 06
Per engineer per month for unrestricted frontier access. The exact number varies by tool and usage. It is not a meaningful number relative to what you pay that engineer.
Per month including salary, benefits, equipment, office space, management overhead. The frontier model is less than 1% of this number. You spend more on their monitor.
One resignation. One recruiter engagement. That single fee dwarfs what you saved on their tokens for the entire year. And their friend at another company just told them they have unrestricted access.
The pushback was never really about the money. It was about the unfamiliarity. "AI tokens" did not exist as a line item eighteen months ago, and new line items get scrutiny.
The real source of budget friction
Slide 07
Your best engineers know who funds their tools with conviction and who funds them with reluctance. They compare notes. They talk to friends at other companies. They notice who has unrestricted access and who is rationing tokens like it is 2023.
You will not see it in your attrition dashboard. You will see it in a resignation letter. And the recruiter bill to replace that person will dwarf what you saved on their tokens for the entire year.
Start with your top twenty engineers. Unrestricted frontier access. Sixty days. Measure the output delta. If it does not justify expanding, pull it back.
That is a decision you can make this week. Not next quarter. Not after the vendor evaluation. Not after the security review that has been sitting in someone's queue for four months.
Slide 08
You built something that let engineers sit down, start a session, and look up four hours later having shipped something real. Not a ticket. Not a story point. Something they are proud of. A lot of us had lost that.
Token limits. Model pickers. Usage dashboards. Smart routing that is not smart enough. Every one of these is an interruption to the thing your product exists to create. You are selling flow and billing for friction.
Friction is the one thing your product cannot survive. The whole reason engineers pay you is flow. The moment you interrupt that, you are selling a car with a governor on the engine. And someone else will sell the same car without one.
You gave me back the joy of building software. Now stop taking it away forty-seven times a day with token limits and model pickers.
The engineer's actual experience
Slide 09
Slide 10
No model picker. No version number. No token anxiety. Just the agent at full strength, all day, every day. That is the product engineers want. That is the investment leaders should make. Everything else is noise.
The companies that figure this out first will have the best engineers, the fastest shipping velocity, and the strongest retention. The ones that do not will be explaining their tiered access model to a room of people who already have offers somewhere else.