AI Engineering Playbook

Database › State of AI in engineering

State of AI in engineering

As of 2026-10-25 53 companies 151 dated sources, newest 2026-10-22 53 of 53 companies have a 2026 source

What the companies in this database have publicly said about AI in their engineering orgs, boiled down into six patterns. Everything below is pulled from the same sourced records as the company pages, and it is regenerated every time the data is rebuilt, so counts and examples change as companies publish new figures. Each point links to the company page and to the dated source behind it.

Read this first: most figures here are reported by the companies themselves or by the vendors that sell the tools, not independently measured. Definitions differ (share of new code, merged code, changes or PRs). See the methodology for how sources are typed and dated, and check each source date, because policies change.

1. AI now writes most of the code at the leaders

20 of 53 companies have published a share-of-code or AI-authored-code figure (full list). The highest current figures:

2. Few companies formally mandate it

Of 53 companies, 6 have a reported mandate (Coinbase, Intercom, Microsoft, Nvidia, Shopify, Zapier), 1 explicitly says use is not forced (Ramp), and 46 are unclear or not reported. Most encourage use and track adoption instead (mandate list).

Duolingo is counted as unclear because it walked back its performance-review requirement. Nvidia's "mandate" wording comes from a case study by Cursor, the vendor that sells the tool.

3. Review is the bottleneck, and the main guardrail

20 of 53 companies describe code review as part of how they control AI-written code. The common pattern is a required human reviewer, often with an AI reviewer first.

4. Coding agents are replacing autocomplete

Companies increasingly publish internal agent platforms that open pull requests on their own, rather than editor autocomplete.

Most common named tools across the database (number of companies): Claude Code (16), Cursor (12), GitHub Copilot (9), Claude (4), Codex (4). Tool lists are on each company page.

5. Org shape is changing

Flatter orgs, fewer pure managers, and less junior hiring show up alongside the AI push. Companies rarely attribute cuts to AI alone, so read the sources.

6. Caveat: most results are self-reported

Of 105 current numeric metric and result claims in the database, 44 rest only on the company's own words (blog, memo or executive post), 21 rest only on vendor-written sources, and 40 cite press coverage, which itself usually relays company statements, such as earnings calls. Across all 151 sources: 20 primary, 38 company, 76 press, 17 vendor. Companies with at least one vendor-written source: Anthropic, Atlassian, Box, Datadog, GitHub, IBM, Mercado Libre, Netflix, Nvidia, OpenAI, Salesforce, ServiceNow, Tencent, Twilio.

Vendor-reported figures, one per company:

Treat any single number as a claim, not a benchmark. Methodology: how sources are typed, dated and verified.