GitLab's AI Strategy: How the All-Remote DevSecOps Leader Listens to Millions of Users in 2026

TL;DR#

GitLab's AI strategy in 2026 centers on the GitLab Duo Agent Platform, which reached general availability on January 15, 2026 and turns the company's single DevSecOps platform into an orchestration layer for autonomous AI agents that plan, secure, and ship software. GitLab crossed $1 billion in annual recurring revenue in fiscal 2026 on $955.2 million in revenue (up 26% year over year), serving more than 10,682 base customers and over half of the Fortune 100. The harder strategic problem sits underneath the product: GitLab is the world's largest all-remote company, with roughly 2,500 team members across 65-plus countries and an estimated 30 million-plus registered users on a platform built around an open-core, MIT-licensed community. A user base that large and that distributed cannot be understood through quarterly surveys — developers, the most survey-fatigued audience in software, ignore them. The lesson for any product-led or developer-first company is that listening has to scale the way the product does, which is why conversational AI research has become the only viable instrument for capturing the "why" behind developer behavior at GitLab's scale. Perspective AI gives product and developer-relations teams that instrument: AI-led interviews that run with thousands of users at once and probe like a researcher would.

Who GitLab Is and Why Its Scale Matters#

GitLab is a publicly traded DevSecOps platform (NASDAQ: GTLB) that consolidates planning, source control, CI/CD, security scanning, and operations into one application. What makes its customer-understanding challenge unusual is the sheer breadth of who uses it. GitLab reports more than 30 million registered users worldwide and roughly 1 million active licensed users, sitting on top of an open-core codebase published under an MIT license that draws contributions from more than 3,300 people across 60-plus countries.

That community is not a marketing footnote — it is the product's distribution engine and its feedback loop. The free tier, the self-managed deployments behind corporate firewalls, and the SaaS instances together form an audience that spans solo open-source maintainers and Fortune 100 platform teams. In fiscal 2026, GitLab's $100,000-plus ARR cohort grew to 1,456 customers (about 75% of ARR), and it counted more than 155 customers above $1 million in ARR, according to GitLab's fiscal 2026 results. A revenue base concentrated in large enterprises, layered on top of a free user base counted in the tens of millions, is a research nightmare for any team relying on traditional survey tooling.

This is the same scaling-the-listening problem we have traced across other category leaders, from the HubSpot AI customer research playbook for a $30B CRM leader to how Stripe runs AI customer research across 4 million businesses. GitLab's version is harder still, because its users are developers.

GitLab's AI Strategy in 2026#

GitLab's AI strategy in 2026 is to make AI the connective tissue of the entire software development lifecycle rather than a bolt-on coding assistant. The centerpiece is the GitLab Duo Agent Platform, which the company brought to general availability on January 15, 2026 — a move it framed as a shift "from individual coding assistance to AI agents that operate across planning, security, delivery, and operations," as GitLab announced in its investor materials.

Three design choices in that platform reveal how GitLab thinks about scale:

• Orchestration over autocomplete. The Duo Agent Platform deploys specialized agents — a Planner Agent for work orchestration, a Security Analyst Agent for vulnerability triage — and an Agentic Chat that uses multi-step reasoning across issues, merge requests, pipelines, and security findings. GitLab is building "AI as a platform," not a single assistant.
• Group-level economics, not per-seat. Duo credits are allocated at the group level and sold in monthly blocks rather than assigned per user, with a usage dashboard showing which agents and flows consume credits. That billing model only makes sense for an organization that expects AI usage to spread across thousands of contributors who never asked for a seat.
• Governance built for distrust. Through the AI Catalog, admins publish approved agent configurations, choose which model vendors power Duo workflows, and exclude sensitive files from AI context, with flat-rate code reviews and free-tier AI access added in April 2026 per InfoQ's reporting.

Every one of those choices assumes a vast, heterogeneous user base. None of them tells GitLab why a developer adopted an agent, abandoned it, or quietly routed around it. That gap is where customer research comes in.

Listening to Millions of Developers at Scale#

Listening to GitLab's user base at scale means collecting qualitative signal from tens of millions of people who are scattered across time zones, deployment models, and trust levels — a problem GitLab already solved once for its own workforce. As the world's largest all-remote company, GitLab runs roughly 2,500 team members in 65-plus countries with zero offices, and documents how it operates in a public handbook of more than 2,000 pages. McKinsey profiled GitLab as a remote-first model precisely because the company turned asynchronous, written, low-friction communication into an operating system.

The customer-research implication is direct: a company that runs itself asynchronously across continents needs research methods that work the same way. Synchronous, scheduled, moderator-led research — the focus group, the 45-minute Zoom interview — does not scale to an audience of 30 million, and it does not respect a developer's calendar. What scales is asynchronous, always-on conversation that a user can pick up at 2 a.m. in their own time zone and finish in their own words.

This is the same scaling principle behind customer research at scale and why the sample-size problem is finally solvable and the broader shift documented in the 2026 state of customer research. For developer-first companies, the method has to be conversational, because the audience treats forms with open contempt.

Why Surveys Fail for Developer Audiences#

Surveys fail for developer audiences because developers are simultaneously the most surveyed and the most survey-averse professionals in software, and the rating-scale format strips out exactly the technical reasoning that makes their feedback useful. The industry-wide collapse is well documented: survey requests have risen sharply since 2020 while completion rates have fallen toward the 12–18% range, with a large share of respondents abandoning surveys partway through.

Developers compound the problem in three specific ways:

1. They detect and resent low-effort instrumentation. A five-point satisfaction scale asks an engineer who just fought a flaky CI pipeline to flatten a detailed technical complaint into a single number. The signal that matters — the specific failure mode, the workaround they built, the config they disabled — never gets captured.
2. Their context is irreducible. The most valuable developer feedback is conditional: "It works fine until the monorepo crosses a certain size," or "I trust the agent for boilerplate but never for security-sensitive code." Forms cannot follow up on "it depends," and developer feedback is almost entirely "it depends."
3. Self-selection skews everything. Even the best developer surveys are self-selected samples. Stack Overflow's 2025 Developer Survey gathered responses from more than 49,000 developers across 177 countries — a strong dataset, yet still a self-selected one — and surfaced that while 84% of developers now use or plan to use AI tools, 46% said they do not trust the accuracy of AI output, up from 31% the prior year per ADTmag's coverage. A 15-point jump in distrust is the kind of finding a rating scale can flag but never explain. Why don't they trust it? On which tasks? After which specific failure? Those answers live in conversation, not in a dropdown.

For a company shipping autonomous agents into developers' workflows, the "why don't they trust it" question is the entire ballgame — and it is the one surveys are structurally worst at answering. The case for replacing the survey layer entirely is laid out in why conversations beat surveys for real customer research and the conversational method that captures the why behind the score.

The Lesson for Product-Led and Developer-First Companies#

The lesson from GitLab is that listening must scale the same way the product does — automatically, asynchronously, and conversationally — or it does not scale at all. GitLab's own playbook proves the point three times over: it scaled its workforce asynchronously, it scaled its product through an open-core community, and it is now scaling its AI through agent orchestration rather than per-seat assistants. A research function that still relies on scheduling interviews or blasting NPS emails is the one piece of the operation that refuses to scale with everything else.

The same dynamic shows up in peer companies pursuing this strategy. Datadog's AI customer research strategy as a $40B observability leader and Asana's customer-research-driven roadmap as a work-management leader both rest on understanding usage at a scale no panel can match, as does ClickUp's all-in-one productivity strategy across 10 million users. The forward-looking read on where this goes for every category leader is mapped in AI conversations at scale: the 2026 state of the category.

This is also why the strongest product analytics on the market are not enough on their own. Behavioral data — the kind a company instruments across millions of sessions — tells you what developers did; it cannot tell you why, a gap explored in Amplitude's AI strategy of pairing behavioral data with customer voice. The "why" layer has to come from talking to people. The same logic drives the conversation-data thesis in Gong's AI strategy for turning conversations into product decisions and Rippling's AI strategy for compounding product velocity with customer conversations.

How Any Team Can Capture Conversational Insight at Developer Scale#

Any product or developer-relations team can capture conversational insight at GitLab-style scale by replacing the survey layer with AI-led interviews that run with thousands of users in parallel. Perspective AI is built for exactly this: instead of a static form, an AI interviewer asks a developer about a workflow, listens to the answer, and follows up on the vague or conditional parts — "you said the agent is fine for boilerplate; what made you turn it off for security work?" — the way a skilled researcher would, across thousands of conversations at once.

A practical path for a developer-first team looks like this:

• Start with one high-stakes question. Pick a single decision — AI agent adoption, a pricing change, a deprecated feature — and run a focused user research interview or customer interview instead of a survey. Both product teams and CX teams can launch one without a dedicated researcher.
• Embed it where developers already are. Use a conversational feature-request intake or user-feedback flow inside the product or docs so feedback is asynchronous and in-the-moment, not a scheduled interruption.
• Validate before you ship. Run roadmap validation or a product-market-fit study with a representative cut of free-tier and enterprise users to pressure-test an agent feature before GA.
• Make research continuous, not quarterly. Spin up new studies on a rolling cadence so the listening cycle matches the release cadence. You can start a new study in minutes, and the AI interviewer agent handles the conversations while a concierge agent routes intake.

Because the AI interviewer runs every conversation in parallel and synthesizes transcripts automatically, a single product manager can field a study with a developer sample large enough to be representative — without booking a single calendar slot. See pricing or compare approaches to scope it for your team.

Frequently Asked Questions#

What is GitLab's AI strategy in 2026?#

GitLab's AI strategy in 2026 is to embed AI across the entire DevSecOps lifecycle through the GitLab Duo Agent Platform, which became generally available on January 15, 2026. Rather than offering a standalone coding assistant, GitLab treats AI as an orchestration layer of specialized agents — for planning, security, and delivery — governed centrally through an AI Catalog and billed by group-level credits rather than per seat.

How big is GitLab's user base?#

GitLab reports more than 30 million registered users worldwide and roughly 1 million active licensed users. On the business side, GitLab crossed $1 billion in annual recurring revenue in fiscal 2026 with more than 10,682 base customers, over half of the Fortune 100, and 1,456 customers paying above $100,000 in ARR. Its open-core codebase draws contributions from more than 3,300 people across 60-plus countries.

Why do surveys fail for developer audiences?#

Surveys fail for developer audiences because developers are heavily over-surveyed, resent low-effort rating scales, and give feedback that is highly conditional and technical. Industry completion rates have fallen toward the 12–18% range, and developer surveys are self-selected samples. A five-point scale cannot capture the specific failure mode or workaround behind a developer's frustration, which is exactly the signal a product team needs.

Why is GitLab being all-remote relevant to customer research?#

GitLab's all-remote model is relevant because it proves a company can operate at global scale through asynchronous, low-friction communication — the same property good customer research needs. With about 2,500 team members across 65-plus countries and no offices, GitLab already runs on async-first methods. Research that depends on scheduled, synchronous interviews cannot match that operating model or reach an audience of tens of millions.

Can conversational AI research replace developer surveys at scale?#

Yes, conversational AI research can replace developer surveys at scale by running AI-led interviews with thousands of users in parallel and following up on vague or conditional answers automatically. This captures the technical reasoning and context that rating scales strip out, while still reaching a sample large enough to be representative. Tools like Perspective AI let a single product manager field a study without scheduling any individual interviews.

Conclusion#

GitLab's AI strategy in 2026 is a case study in scaling everything at once: the product scales through the Duo Agent Platform, the company scales through an all-remote operating model, and the business scales past $1 billion in ARR — all on top of a 30 million-plus developer user base that no survey could ever truly hear. The piece that has to keep up is listening. For any product-led or developer-first company, the conclusion is the same one GitLab's own playbook points to: understanding your users has to be automatic, asynchronous, and conversational, or it falls behind everything else you build. Perspective AI gives product, research, and developer-relations teams that instrument — AI interviews that scale to thousands of conversations and capture the "why" a dropdown never will. Start a study and listen to your users the way GitLab listens to its developers: at scale, in their own words.