Anthropic's own data: engineers use AI on 60% of their work but still won't 'fully delegate' more than a fifth of it

Anthropic's 2026 Agentic Coding Trends Report — drawn from its own internal engineering data and customer deployments — finds developers use AI in roughly 60% of their work but report being able to fully delegate only 0–20% of tasks, and that about 27% of AI-assisted work is net-new output that wouldn't have happened otherwise.

Anthropic has published its 2026 Agentic Coding Trends Report, and the headline number isn’t the one you’d expect from a company selling coding agents. Engineers report using AI in roughly 60% of their work, but they can “fully delegate” — hand off without active supervision — only 0–20% of tasks. The gap between those two figures is the report’s central finding, and it cuts against a year of “AI writes most of the code now” framing: the tools are doing more, but humans aren’t stepping back proportionally. They’re reviewing, directing, and validating instead of writing.

Where the productivity actually shows up

The report’s more useful number for anyone commissioning software is this: about 27% of AI-assisted engineering work is work that wouldn’t have happened otherwise — scaling projects, building nice-to-have internal tools, exploratory work that wasn’t cost-effective to do manually before. That’s a different claim than “faster delivery of the same scope.” It’s more scope, more polish, more of the backlog getting cleared, for the same budget. Anthropic backs this with customer case studies: TELUS shipped code 30% faster and logged over 500,000 engineer-hours saved; CRED doubled its execution speed on production fintech systems; Rakuten had Claude Code complete a complex extraction task across a 12.5-million-line codebase in seven hours of autonomous work, matching a reference implementation to 99.9% numerical accuracy; an Augment Code customer finished a project their own CTO had estimated at four to eight months in two weeks.

The catch is in the fine print, not the marketing

None of this is “hand the project to an AI and check back later.” Anthropic’s own researchers describe engineers developing intuitions over time for what’s safe to delegate — typically tasks that are easily verifiable or low-stakes — while conceptually difficult, design-dependent work stays firmly human. That’s consistent with what serious engineering teams have been saying all year: the tooling has changed faster than the amount of judgement required to use it well.

So what

If a development partner’s pitch is “AI means it’s basically free and instant,” that’s not what the company building the leading coding models is actually reporting. If the pitch is “AI lets us ship more — more edge cases handled, more of your backlog cleared, more iterations — inside the same budget, with an engineer still reviewing every line that matters,” that matches the data. Ask any partner quoting an AI-accelerated build which of those two stories they’re actually selling. See our AI-assisted development approach for how we scope projects around that distinction, or get in touch to talk through what it means for your build specifically.

← All research

More research.

GitHub secret scanning cuts false alarms for AWS, Google and Alibaba cloud keys and adds a SpaceX detector

GitHub's 5 November 2026 secret scanning update tightens detectors for AWS, Google Cloud and Alibaba keys, relevant to teams building software with AI coding tools that can paste credentials into code.

Read ›

GitHub's Copilot code review can now be managed by API, with review effort levels and AI credit costs shown

GitHub's 6 November 2026 update lets organisations manage Copilot code review by API and see the review effort and AI credits each AI code review used, a sign AI-assisted development is becoming auditable.

Read ›

GitHub's Copilot impact dashboard now counts chat and VS Code agent use, and drops the 'passive users' label

On 6 November 2026 GitHub added Copilot Chat and the VS Code Agents window to its Copilot impact dashboard, bringing it to nine tracked experiences, and renamed 'Passive users' to 'Chat-first users'.

Read ›

Claude Code 2.1.319: a setting that denies permission prompts so AI agent runs don't stall

Claude Code 2.1.319 (6 November 2026) adds a permissions.denyInsteadOfPrompting setting and moves long MCP tool calls to the background, a sign AI-assisted development is being built for unattended runs.

Read ›

GitHub now reports how often developers act on Copilot code review comments

On 5 November 2026 GitHub added code review metrics, including an addressed rate, to the Copilot usage metrics API. For AI software development buyers, it is the first first-party number for how much AI review feedback is acted on.

Read ›

Dependabot can now keep AI agent skills and prompts up to date

GitHub's Dependabot gained Agent Package Manager (APM) support on 3 November 2026, treating AI agent skills, prompts and instructions as dependencies that need updating like any other software.

Read ›

Start a project

Building with AI? Let's talk.

Tell us what you are building. We'll give you an honest view of the right approach.

Start a projectSee our work