How do you handle estimates in the modern era?
I have always struggled with estimation. For most of my nine-year career, I was a serial underestimator, and I’ve only really gotten the hang of it in the last few years. I’ve relied on the standard tricks: multiply the initial estimate by 1.5, account for the percentage of engineering time that will be lost to other work, and add a buffer for testing.
Now, though? I have no idea what to do. Things that used to take weeks now take days. My team was just assigned a “half-year” project. It has a ton of scope: we’re building an entire backend system and a new internal app.
We spent a few weeks on design, and then pressure came from above to have the core of it ready in a month and a half. My first instinct was to push back violently, but instead, I decided to spend a day putting together a plan for agents to implement a simple demo. I had AI do some market research for the app and examine potential features and UX. On Friday, I gave the design, constraints, and market research to an ultracode agent with a single goal: give me a working demo backed by realistic synthetic data, with fully functional plumbing.
It’s already done, so now I’m having it add extras…
Honestly, it’s far beyond my expectations, and I only had to steer it three times over the course of two days. It’s obviously not ready for prime time. There’s still security, performance, God knows what new requirements will come in, and the work of pulling in non-synthetic data once everything is secured. The point, though, is that this was a very nontrivial amount of work, work we estimated would take a team of three more than two weeks.
To make a long story short, I have no idea how to give estimates anymore when AI continuously surprises me with how quickly and thoroughly it can accomplish things. My team estimated two weeks just for a prototype. On a goddamn whim, I had a more complete version finished in three days. Some of the most time-consuming elements of the job have become trivial, yet other elements are completely unchanged, and it’s hard to know how to weight them.
Does anyone have updated tips or tricks? Leadership is continuously pushing for accelerated timelines, and I honestly think we can meet them. But every instinct I’ve developed over almost a decade in the field screams at me to fall back on the ultra-safe estimates that I know are overestimates in today’s world.