Topic
The practice of measuring software engineering output and efficiency, using metrics like code shipped, task completion time, and cost per feature.
Red Hat's post puts generative AI's productivity gain on the hardest enterprise codebases at 0-10 percent, then offers a pipeline that grounds the model in migration rules a customer's own team has to write and curate.
Reality
CEO Itamar Friedman says the ceiling exists to force someone to answer which automation is worth the money. The bill growing faster is the infrastructure Qodo runs for customers, up roughly 5x year over year.
Reality
An agent drafts the CSV import in an afternoon; the week of making it correct is what remains billable. Three large surveys say what that week is worth. A controlled study says developers get their own hours wrong.
Reality
Daily agent use among engineers climbed to roughly 80% in a year while developer trust in the output fell to 29%. The case that the gap is costing delivery time rests on a randomized trial of 16 people.
Reality
A dev.to post argues that the approve and block labels AI review tools print are uncalibrated text, and proposes fitting sigmoid(alpha + beta * verdict) to a team's own review history so triage gets a threshold it can defend.
Reality
A field study of 4,867 developers recorded 26.08% more tasks completed with GitHub Copilot, while METR timed 16 experienced developers taking 19% longer in repositories they already knew. The two figures count different things.
Reality
A principal engineer built the same EV-charging invoice service twice and timed a coding agent through nine cumulative features on each. The two setups differed in more than layering, and he says so.
Reality
Anthropic's analysis of roughly 400,000 sessions puts about 80 percent of the execution decisions with the agent. The benefit it reports is longer unattended runs between check-ins, counted in actions per turn.
Reality
The step in front of the person got faster. The steps behind it did not, so an afternoon of work aged in review for months. A survey of 4,500 executives in nineteen countries describes the same condition at scale.
Reality
GitClear's Maintainability Gap report measures output and code health on the same 623 million changes. The two move in opposite directions, and the denominators cover different teams.
Reality
A consultant's six-month log across forty shipped Python automations reports build time 38 percent below baseline and client revisions halved, with an explicit warning that React work sits outside the sample.
Reality
Cursor's Projects beta gives each body of work a coordinator agent on its own cloud machine. Its internal deployment shows where the review load lands, and its productivity figures arrive without a baseline.
Reality
Anthropic's usage dashboard reports each team member's email beside their monthly lines of code accepted and exports the list as a CSV, which turns a seat-utilisation tool into something a performance review can reach for.
Reality
Sean Goedecke argues that visible anger gets you excluded from the conversations where the interesting work is handed out, and the awkward part is that his own essay rates those engineers above average at shipping, so no dashboard catches the loss.
Reality
A 90-day instrumented study reports 28% more pull requests and 13 points of test coverage, but the gain lands only after training and workflow embedding, which puts the money on enablement rather than seats.
Reality
Andrew Swerdlow says Roblox spent about six months moving from autocomplete to agents, and that the binding constraint is trust and review infrastructure rather than model choice.
Reality
A theoretical paper assumes error-free, near-free language models and still finds that in two of three deployment scenarios, researchers respond by doing more work with less care.
Reality
One iOS team's own measurement puts the value of agent workflows in the mechanical half of code review. In Swift, letting agents generate freely moved the time the other way.
Reality
The multi-agent pipeline is the part everyone will copy. The measurement, and the categories where acceptance falls to 40.6%, is the part worth reading.
Reality
The reported gain sits in code generation and the reported costs sit in review and post-merge repair. Computed against one unit of work, most teams do not come out ahead.
Reality
Build · August 21, 2026 · 1 publisher
Product · August 21, 2026 · 1 publisher
Product · August 20, 2026 · 1 publisher