Moving from AI adoption to an AI operating model: How Docusign engineering went beyond tool rollout
Nick Lucchesi with Sagnik Nandy · 8 November 2026
Read the originalDocusign spread AI coding tools across more than 1,100 engineers through embedded helpers and shared AI agents, and now judges the effort by delivery and reliability rather than usage.
Docusign, which sells electronic signature and agreement software, gave its engineers AI coding tools. By June 2025, 49% of engineers used them and a reported 18% productivity gain had stalled. In The New Stack, chief technology officer Sagnik Nandy describes how a central team, embedded helpers and shared AI agents brought all of more than 1,100 engineers into active use.
- A six-person Eng AI Champions team set guardrails and ran workshops. About 75 AI Multipliers inside product teams then spread the practices and reported problems back, and active use reached 93% within two months.
- Engineers build AI agents for recurring tasks on an internal platform called Krew. Every agent is visible to all teams with its usage, and any team can copy and extend it.
- One overnight agent searches error logs and leaves an engineer a proposed fix as a pull request each morning. Another changed 10 of 12 configuration files correctly in about 15 minutes, against roughly a week by hand.
- No agent may merge a pull request or take an irreversible action without a person's approval, and the engineer who approves a pull request answers for it.
An engineering organisation adopting AI agents can pair a small central team with engineers embedded in each team. It can then judge the effort by delivery time, reliability and rework rather than usage. This is a sponsored post in which Docusign describes its own programme. The reported 52% rise in lines of code per developer does not by itself show faster delivery, as the article notes.
Make every shared AI agent visible to all teams with its usage count, so that teams copy and extend a working agent before building their own.
Require a named person's approval before any AI agent merges a pull request or takes an irreversible action.
Derived by Working Surface from the article; more in the Playbook. Source line: “If something is good, you can try to teach 1,000 other people to do it, or you can make it easy for 1,000 other people to inherit it.”
9 November 2026: as AI agents remove the slow steps that once gave teams time to question the work, teams are writing down the owners, measures, rules and limits those steps used to supply.