How We Drove AI Adoption Across Docusign Engineering Without Mandating It

Bharat Rele, senior manager at Docusign, and Nabeel Shahzad, director of engineering at Docusign, also contributed to this blog.

Docusign is in the business of building AI products. Our Intelligent Agreement Management (IAM) platform, powered by our AI engine Docusign Iris, brings AI to millions of contracts every month. Increasingly, we've been using AI to help build those products.

By mid-2024, AI coding tools were available at Docusign, but like a lot of engineering orgs, we weren't really using them. Adoption sat in the low double digits. The question was how to get engineers to actually use them day to day, in a way that would stick. 

Today, 100% of our 1,000+ engineers are onboarded to and actively using AI coding tools, and around 75% of our new code is AI-assisted, with every line still reviewed and approved by a human engineer before it ships. 

Getting here required moving through a deliberate, structured evolution from trying AI as an isolated tool, to using it via a viral peer network, to trusting it to manage end-to-end engineering workflows. Our Eng AI Champions program was the key. This peer-led model took us from low-double-digit adoption to near-universal in a little over a year, and it taught us a lot about how our engineers actually work.

Here’s how we scaled AI across Docusign engineering, the architectural framework we built around it, and the metrics showing early impact.

Docusign's AI coding journey

Phase 1: The pilot and the paradox of access (June 2024 – June 2025)

Focus: Ground-level experimentation and identifying adoption friction.

Following a successful pilot program from December 2023 to June 2024, which enabled GitHub Copilot for roughly 20% of our engineering organization, the technical promise was undeniable. This early success prompted us to open access to everyone, initiating a company-wide rollout in June 2024. 

But as the wider rollout progressed, we ran into a ceiling. Access alone got us partway, but it stalled.

• Adoption reached 49%: Nearly half of our Docusign engineers began actively using AI coding tools in their daily workflows.

• Developer productivity increased by 18%: We saw an initial lift in average productivity across the entire company by using these AI tools, but this momentum eventually stalled.

Both adoption and productivity had plateaued. The AI coding tools were objectively good, and our engineers were curious, but we identified clear friction points: dedicated time to experiment, guidance tailored specifically to our unique codebase, and a collaborative structure to lean on when prompting didn't work exactly like the marketing demos promised.

Phase 2: Building the web of network (June 2025 – Nov 2025)

Focus: Organizing the grassroots from AI Champions to AI Multipliers.

We realized that top-down corporate mandates have limits in engineering cultures while peer-to-peer enablement wins. To capture the organic energy of passionate developers experimenting across different teams, we formalized a central task force: the Eng AI Champions program.

The initial mandate of the six-person AI Champions team was to establish structural guardrails, curate engineering best practices, design workshops, and architect a strategy to weave AI across the entire Software Development Lifecycle (SDLC). By October 2025, this focus drove adoption up to 65%.

But a centralized team of six couldn't scale deep into every daily standup. To cross the chasm to full adoption, we introduced the AI Multiplier program: working engineers embedded directly within individual development teams who happen to be early adopters and exceptional peer mentors.

Crossing the chasm to full adoption

This created a highly effective bi-directional feedback loop:

• Top-down support: The AI Champions work closely with the Multipliers to share best practices, coach them, and provide platform updates.

• Ground-level enablement: In turn, the Multipliers work side-by-side with their direct teammates, helping peers navigate their first week with an AI tool, run localized workshops, and collect real-time friction points.

• Bottom-up feedback: Those friction points are fed directly back to the Champions to refine the overarching company strategy.

The impact: developer productivity and democratization

The impact was immediate: within just two months of launching the Multipliers, our active user rate skyrocketed to 93%, and lines of code updated per developer had increased roughly 52% compared to where we started.

The immediate result of this network wasn't just tool adoption, it was skill democratization. We watched a backend team with zero frontend experience build a functional frontend widget using AI. A designer with no coding background built a functional prompt library and frontend widget using Claude Code. AI began fundamentally lowering the barrier to execution, helping engineers confidently build in areas they wouldn't have tackled alone.

Phase 3: Moving beyond code to workflow automation (Dec 2025 – Present)

Focus: Scaling AI agents across operations, code review, and living knowledge.

Once AI became a natural tool for writing code across Docusign’s engineers, we shifted our focus from optimizing individual code-completion tasks to automating entire end-to-end engineering workflows. Today, we look at the SDLC holistically, deploying specialized internal AI agents we built to handle the heavy lifting of code review, system health, and on-call investigations:

AI for coding: from assistance to autonomy

To handle the long tail of well-defined, repetitive tasks that consume developer time but require minimal judgment, we also built an autonomous coding agent internally dubbed our "coding elf." An engineer describes a change in plain English via Slack, Jira, or directly in-app; the elf autonomously writes the code, executes tests, and submits a pull request for human review. It even monitors its own PRs, reads reviewer feedback, and pushes updated iterations automatically.

The elf's simplified workflow for turning natural language descriptions into code

Docusign's AI coding tool

In 4+ months, the share of merged PRs generated by the elf has grown from under 1% to over 8%. A legacy infrastructure task that previously required an entire week of an engineer's time — modifying more than 10 configuration files and cross-referencing values across them — was completed by the elf in 15 minutes with 90% accuracy on the first pass.

We wrote about how we built our AI coding elf here.

We also built our own internal AI-assisted code review tool, which now processes more than 3,000 pull requests per day across 1,600+ repositories. It provides quantitative evidence of catching certain architectural and security flaws earlier in the review cycle than human reviewers would, shortening overall turnaround times.

Context is king: living documentation via MCP

AI needs real-time context of the codebase to be genuinely useful. We built DocuWiki, an internal tool with a Model Context Protocol (MCP) server that lets engineers securely "chat" with their codebase — turning any repository into living, searchable documentation.

AI-assisted observability for on-call teams

Most monitoring tools are good at detecting that something broke; the harder problem is telling an on-call engineer at 2am whether the change actually matters, since traditional dashboards leave humans to piece together context across services. We built our own internal AI observability platform to close that gap. It correlates live telemetry data across more than 400 services. Out of 16,000+ active monitors, it has surfaced 800 anomalies while filtering the noise down to roughly 80 actionable alerts.

This 10-to-1 signal-to-noise ratio drastically reduces alert fatigue and accelerates automated root-cause analysis. It has already completely replaced our legacy monitoring tool for CLM, with the ultimate vision of predicting and resolving system issues before they touch a customer.

AI-powered issue investigation

Our in-house AI agent autonomously investigates issues end-to-end. It queries live telemetry, analyzes source code, and posts root-cause analysis and recommended fixes to Jira or Slack in minutes. Acting like an engineer, it then hands off implementation to our AI coding tool, which generates the code changes needed to resolve customer-reported bugs.

The skill marketplace platform

So that an efficiency gain in one team benefits the entire global engineering organization, we're deploying a Skill Marketplace platform, where engineers can package, publish, and discover successful AI prompts, agent configurations, and custom automated skills company-wide.

The impact: hardening workflows and deepening gains 

Migration and conversion projects, where the work is well-defined but repetitive, are proving to be a productivity boon for our engineers. 

Across Phase 3, lines of code updated per developer increased a further 32%, pushing our total growth to a 101% increase compared to our original baseline. While Phase 3's incremental growth is more modest than Phase 2, it reflects our transition from driving initial tool adoption to hardening our daily engineering workflows. 

Adoption and productivity gains from AI coding over time

What’s next

We think of our Eng AI Champions journey in four phases: Try, Use, Trust, Rely. We've moved past "try," and the vast majority of our engineers are actively using AI tools daily. We've built the foundation: tools deployed, engineers onboarded, early wins emerging. Now the focus is proving AI's value on well-scoped tasks and building the confidence to rely on it more broadly.

We don't have all the answers yet. AI capabilities are evolving quickly, and so is our understanding of where they add the most value across our engineering teams so we can serve our customers better and faster. But we're confident in the direction. -

Is your engineering team also leveraging AI to work more efficiently? If so, I’d love to hear about your approach, so feel free to connect with me on LinkedIn opens in a new tab. And if you want to join our team of engineers using AI to shape the future of AI-powered contracts, explore our open roles here.