Tools / 2026 / Live in the browser
Autopilot Ops
Ranks engineering work by value over effort and packs the week's budget for the most total value, with the same planner in Python and in the browser.
01
The problem
Taking the best-scoring tasks first until the hours run out is Dantzig's greedy rule, and it leaves value on the table: with 10 hours, one 6 hour task with the best score beats two 5 hour tasks that are worth more together.
02
How I approached it
One Python core behind a CLI and a FastAPI endpoint, plus a line-for-line browser port the page falls back to when the API is not there. The weekly plan is an exact 0/1 knapsack solved by dynamic programming over hundredths of effort, rounded up so a plan never overruns, with a strict tie rule so both ports pick the same plan. Inputs reject NaN, infinite and negative numbers, and confidence reads 80 as 80% the way RICE writes it.
03
The outcome
On the sample set with a 5 hour budget, score order plans 20.5 of value and the packer plans 23.3. Both ports are checked against brute force on 60 random task sets, and the page says why each task was left out.
Related work
A dependency-free CLI that scaffolds a runbook from a project's own package.json and checks READMEs and runbooks by their real headings, with a score and a CI gate.
A focus timer that keeps time when you switch tabs, rings when time's up, and shows where your week went. Ember, a little spark mascot, keeps you company.
Scores a backlog by leverage, value per unit of effort discounted by confidence, then packs the week against a real budget and tells you what it left out.