Skip to content
Boxed Online, home

Work / Evidence

Quoted versusdelivered.

The full log, for anyone who needs to check before they commit budget. Each row is a build, the estimate written before it started, and what was delivered. Nothing here is rounded in our favour.

Method

How the numbers were made.

  1. Estimated before the build

    Each estimate comes from a written plan, produced before any code, sized as a conventional team would size it. The plans are still in the repositories, dated.

  2. What delivered means

    Built, tested and working against acceptance criteria. It does not mean production-hardened. Penetration tests, performance runs and identity set-up follow, and are part of a build engagement.

  3. Why team-weeks

    A team-week is the effort a conventional team would put in. Quoting it that way keeps the comparison fair and stops the claim sliding into a boast about our own speed.

At a glance

The gap, to scale.

Each bar is one build: the estimate written before it started, shrinking to the time it took to deliver. The figures are the same as the log below.

Condexa rule designer52-63 developer-days for two developers

9-10 weeksOne evening

Healthcare scheduling platformAbout 560 person-days across three phases

29 team-weeks2 days

KashPots refactorRoughly 5.5 working weeks for one engineer

5.5 weeks4 hours

Multi-tenant access-control framework248 person-days for 4.5 people

14 weeks5 days

Technology budget and approvals appSliced plan for a team of 4.5

15.5 team-weeks2 days

The log

Every build, measured.

Quoted versus delivered for every measured build. Estimates in weeks of estimated effort for a conventional team; delivered means built and tested.
BuildEstimated before the buildDeliveredEvidence
Condexa rule designer9-10 weeks52-63 developer-days for two developers, written in the plan before the buildOne eveningEight build phases committed in about 3.5 hours; merged the next day after UX fixesPlan and commit history in the repository
Healthcare scheduling platformClient work, anonymised29 team-weeksAbout 560 person-days across three phases, sized for a team of about five2 daysAll three phases built and passing acceptance tests; phase two in about an hourPhase plans and commit history
KashPots refactor5.5 weeks"Roughly 5.5 working weeks for one engineer"4 hours57 of 64 planned tasks (89%) complete in one morningRefactor plan and branch commits
Multi-tenant access-control frameworkClient work, anonymised14 weeks248 person-days for 4.5 people5 daysWorking packages in a day; a remediated alpha with 822 automated tests in fivePlan, package feed and 191 commits
Technology budget and approvals appClient work, anonymised15.5 team-weeksSliced plan for a team of 4.52 days14.5 team-weeks of slices built and workingSlice plan and project notes
PrezzyListNot estimatedBuilt straight from a product concept3 daysIdea to live platform with CI/CD, observability and a full onboarding epic26 commits, 28-30 September 2026

Sources: dated plans, git history and project notes. Client builds are anonymised with permission.

What the research says

The honest picture.

Independent evidence on AI and developer speed is mixed. We would rather show you that than hide it.
METR, 2025
In a randomised trial, experienced open-source developers were 19% slower when they used AI tools, while believing they were faster.
Read the source (METR, 2025, opens in a new tab)
Google DORA, 2025
AI amplifies whatever engineering practice is already there. Strong teams get stronger; weak practice gets louder.
Read the source (Google DORA, 2025, opens in a new tab)

So the speed does not come from the tool. It comes from the process and the architecture discipline around it. That is what you are buying.

Want the same result on your problem?

In 30 minutes you'll know whether to fix it, buy it, build it or leave it. No pitch deck.

We reply within 2 business days.