Are we typists? Or are we engineers?

$2.2 million.

That’s what my 2026 (so far) would have cost in 2022. Instead it cost about $2,200 in AI subscriptions and tokens. I haven’t typed a line of code since January 1.

In that time I’ve shipped six internal products at Upstatement/SYLVAIN, five of them from scratch: a finance and operations platform with 27 utility apps, a demand-forecasting app, a proposals platform, a reporting suite, a PDF generator, and dozens of new models and improvements to our data warehouse.

About 230,000 lines of code, 44,000 lines of tests, 5,400 commits, and 750 pull requests. AI did the typing. I directed and reviewed it.

The $2.2M comes from pricing the work at 2022 staff costs: about 10 engineers plus product, design, and management support for nine months (on track for $3M by year end).

Would we have spent $2–3 million on internal tools? Of course not. These apps would have lived on the “we should do this someday” list forever.

Because in 2022, software came as a bundle: judgment and typing. The judgment is deciding what to build, nailing the core use cases, making the big engineering calls, and deciding how the products fit together. The typing is turning all of that into 230,000 lines. You couldn’t hire one without the other, even though typing was most of the hours.

AI engineering split the bundle. When typing costs are basically pennies, judgment stops being rationed. In 2022, exploration and refinement had to justify their build cost. Want to try a second layout? That’s a sprint. Rethink how two apps divide the work? Maybe next quarter. So you settled early, did your best, and called it scoping.

Now changing your mind costs an afternoon (or less). I build the second version, throw away the first, and try a third.

The part I didn’t expect is what it did to how I hear feedback. For most of my career, a bug report or a feature request was a new cost. So we got good at defending against them: “working as intended,” “let’s put it in the backlog,” “what problem are you really trying to solve?” Some of that was wisdom. A lot of it was artful deflection to save $$$.

Now when a stakeholder or user has a better idea, I get to say “hell yeah, let’s do it.” You need to browse by month? Add notes to a record? Create a Slack verification loop? Let’s try it and ship tomorrow. I love complaints now: they’re just the next thing to build.

Typing got cheap, and the judgment finally gets the whole day.