Is the AI Code Assistant Market growing now?
In our AI code assistant market deck, you will find everything you need to understand the market
SUMMARY
Yes. The AI Code Assistant Market is growing very fast now, with spending, enterprise deployment and the depth of agent use all rising at the same time.
Developer adoption is already close to saturation. JetBrains found 90% of professional developers using AI coding agents at least weekly, so the next stage of growth will come less from first-time users and more from daily use, longer tasks and multiple agents running in parallel.
Commercial growth is broad rather than concentrated in one winner. GitHub Copilot, Claude Code, Cursor and Cognition are all adding paid enterprise usage while reporting sharply higher subscriber counts or annualized revenue.
Market-share losses do not currently mean business contraction. Cursor reportedly doubled annualized revenue from about $2 billion to $4 billion while its share of workplace adoption in the JetBrains survey fell, which shows how quickly the underlying category is expanding.
Enterprise adoption has moved well beyond small pilots. Goldman Sachs says its 12,000-plus developers use advanced coding agents, Siemens rolled Copilot out to 30,000 developers, and Microsoft says nearly 140,000 organizations now use GitHub Copilot.
The bigger change is what developers are delegating. Coding agents are moving from autocomplete toward multi-hour jobs, repository-wide changes, testing, pull requests and several concurrent tasks, which makes each active user economically much more valuable than an old autocomplete seat.
Productivity evidence is now strong enough to support large deployments, but there is no universal gain. Microsoft researchers found roughly 24% more merged pull requests among adopters, while METR and DORA show that review, integration and task selection can absorb part of the apparent speedup.
Production quality is good enough to be useful, not good enough to trust blindly. Agent-generated code still misses tests, can optimize for what is checked rather than what was requested, and often needs heavier review than human-written pull requests.
Rising usage is also changing the economics of the category. Long-running agents consume far more inference than autocomplete, pushing customer spending upward while creating real gross-margin pressure for vendors.
The core market already looks roughly like a $10 billion annualized enterprise software category, and the competitive fight is getting harder. The open question is no longer whether demand exists, but which vendors can turn that demand into durable margins, distribution advantages and control of the engineering workflow.
This market map, featured in our AI code assistant market deck, highlights top companies and startups in the AI code assistant market
Is the AI Code Assistant Market Growing Now?
What counts as an AI code assistant now?
The AI code assistant market currently covers tools built mainly to help developers write, understand, review, test and maintain software, including GitHub Copilot, Claude Code, Cursor, OpenAI Codex and Cognition's Devin.
The definition has widened quickly. GitHub Copilot originally became popular through code completion, while the products gaining attention these days can inspect repositories, plan changes, edit several files, run commands, test their work and create pull requests. Gartner now calls this category "enterprise AI coding agents" and requires products to support autonomous task execution, context management, human oversight and integrations across the development workflow.
We would keep Lovable, Replit and similar prompt-to-app products next to this market rather than inside the core calculation. They increasingly let founders, designers and other non-developers build software, which makes the wider software-creation opportunity much larger. The same caution applies to ChatGPT, Claude and Gemini: all can write code, but coding represents only part of their usage.
The boundary is getting blurrier. OpenAI recently said about 20% of Codex users were already non-developers, even though Codex started as a software-development product. For this analysis, we focus on spending and usage tied primarily to professional software work.
Are developers still adopting AI coding tools, or has everyone already tried them?
AI code assistant adoption is still growing, but professional developer reach is now so high that the next leg of growth will come mostly from heavier usage rather than millions of untouched developers.
JetBrains' latest Developer Ecosystem Survey, fielded from May through July 2026 across more than 15,000 professional developers, found that 90% were using AI coding agents at work at least weekly and 68% were using them every day. That is already close to universal exposure among professional developers.
The interesting part now is frequency. A developer who used Copilot autocomplete occasionally two years ago and a developer who runs Claude Code for several hours while supervising other agents both count as AI coding users, but their economic value to vendors is completely different.
We therefore expect user penetration to become a weaker growth metric from here. Daily use, number of simultaneous agents, tasks delegated and tokens consumed should tell us much more about where the AI coding market is heading.
If you want more recent data on this point, please see our latest AI code assistant market report.
As this chart shows, and as featured in our AI code assistant market deck, search interest in AI code assistants has increased significantly
Are companies actually paying more for AI coding tools?
Companies are currently paying a lot more for AI coding tools, and the growth shows up across Microsoft, Anthropic, Cursor and Cognition at the same time.
Microsoft reported 4.7 million paid GitHub Copilot subscribers in its fiscal second quarter, up 75% year over year. By the following quarter, Microsoft said enterprise Copilot subscribers had nearly tripled year over year.
Anthropic has seen an even faster ramp with Claude Code. In its latest disclosed figures, Claude Code was above $2.5 billion in run-rate revenue, more than double its level at the start of 2026. Business subscriptions had quadrupled, and enterprise customers were generating more than half of Claude Code revenue.
Cursor gives us another independent example. Forbes reported that Cursor's annualized revenue rose from roughly $2 billion in February to $3 billion in late April and about $4 billion in early June. Cursor also said around 75% of that run rate was coming from businesses, after starting life mainly as an individual developer product.
Cognition is smaller but growing at a similar speed. The company said in May that run-rate revenue had reached $492 million while enterprise use of Devin had increased more than tenfold since the start of the year. TechCrunch later reported that enterprise usage had been growing about 50% month over month for six months.
When four large competitors are simultaneously adding paid users, enterprise accounts and hundreds of millions or billions of dollars in annualized revenue, we are looking at category growth rather than one exceptional company.
Are GitHub Copilot and Cursor actually shrinking?
GitHub Copilot and Cursor are losing developer usage share right now, but both can still grow commercially because the AI coding market is expanding faster than their share is falling.
JetBrains provides a clean view of the rotation. Claude Code went from 18% workplace adoption among professional developers in January to roughly 39% in the May-to-July survey. Codex jumped from 3% to 16%. Over roughly the same period, GitHub Copilot moved from 29% to 21% and Cursor from 18% to 12%.
That sounds bad for Cursor until we compare it with the company's revenue. Cursor's reported annualized revenue doubled from about $2 billion in February to $4 billion in early June while its share in the JetBrains survey was falling. Its enterprise business was growing fast enough to outweigh weaker relative adoption among individual developers.
Microsoft shows a similar pattern from a different angle. Copilot has clearly lost its old position as the default specialized AI coding tool, yet Microsoft's paid subscriber base is still expanding quickly. Share loss and business contraction are two different things when the whole market is moving this fast.
The turnover also tells us something important about developer behavior: loyalty is weak. Developers are willing to switch when another agent gets noticeably better, and they often keep several tools at once.
| AI coding tool | Workplace adoption in January 2026 | Workplace adoption in May–July 2026 | Change |
|---|---|---|---|
| Claude Code | ~18% | ~39% | +21 pts |
| GitHub Copilot | ~29% | ~21% | -8 pts |
| OpenAI Codex | ~3% | ~16% | +13 pts |
| Cursor | ~18% | ~12% | -6 pts |
If you want more recent data on this point, please see our latest AI code assistant market report.
This chart, featured in our AI code assistant market deck, illustrates yearly VC funding for AI code assistant startups
Have big companies moved AI coding agents beyond pilots?
Large companies have moved AI coding agents well beyond pilots, with individual deployments now reaching tens of thousands of developers.
One of the freshest examples comes from Goldman Sachs. In a Business Insider interview published this week, CIO Marco Argenti said all of Goldman's 12,000-plus developers were using advanced agentic AI tools such as Claude and Devin. The firm's problem has already moved on from getting engineers to try agents: Goldman is now teaching those agents its internal engineering standards, security rules and cloud-migration practices.
Siemens followed a different route but reached similar scale. Microsoft said the company adopted the broader GitHub platform after successfully rolling Copilot out to 30,000 developers.
Across the market, Microsoft now says nearly 140,000 organizations use GitHub Copilot and enterprise subscribers have nearly tripled year over year. Cognition counts companies such as Mercedes-Benz, Goldman Sachs, Citi, Dell and Santander among its customers. OpenAI has also disclosed large Codex rollouts, including Sea, where it measured an 87% weekly-active rate among enabled users.
A year or two ago, enterprise adoption often meant a few hundred developers testing a coding assistant. These days, the interesting deployments cover whole engineering organizations.
Are AI coding agents really doing whole software tasks now?
AI coding agents are now taking on multi-hour, end-to-end software tasks, with developers increasingly supervising jobs rather than typing every change themselves.
OpenAI's Codex usage research gives us a useful measure of task depth. By May, 70.2% of sampled individual Codex users had attempted at least one task estimated to represent more than one hour of human work. About 25.6% had attempted a task estimated above eight hours, and more than 10% were managing at least three Codex agents during the same week.
Product design is following the behavior. GitHub now has cloud agents, a CLI agent, multiple concurrent sessions, automated pull-request workflows and enterprise analytics that separately track agent activity. Recent Copilot releases have added agent plugins, subagent task management and more automation around long-running work.
Claude Code has also moved toward greater autonomy. Anthropic recently made its auto mode the default for several paid plans, allowing Claude Code to continue working without requesting approval at every step unless an action looks irreversible, destructive or external to the environment.
Even the place where developers interact with these agents is changing. Slack recently launched Slack Code channels where teams can bring in agents including Claude Code, Devin and GitHub Copilot, inspect the work together and approve changes. The coding agent is gradually becoming a participant in the team's workflow rather than a box beside the text editor.
This chart, featured in our AI code assistant market deck, breaks down Anyshpere’s playbook in AI code assistants
Do AI code assistants actually make developers faster?
Yes, AI code assistants are making many developers faster now, although the measured gain changes dramatically with the task, codebase and team.
A recent study of tens of thousands of Microsoft engineers offers some of the better real-world evidence we have. Researchers studied the early-2026 rollout of Claude Code and GitHub Copilot CLI and estimated that adopters merged roughly 24% more pull requests than they otherwise would have. The effect remained visible across the four-month study window.
METR's evidence shows why we should resist turning that into a universal productivity number. Its well-known randomized study of experienced open-source developers using early-2025 AI tools found that developers actually took 19% longer. More recent METR work using later-generation agents pointed toward productivity benefits in roughly the 4% to 20% range, although the researchers warned that selection effects had become strong because some AI-heavy developers no longer wanted to participate in tasks where AI might be prohibited.
Google's DORA research reaches a similar middle ground from enterprise surveys. More than 80% of technology professionals said AI had increased their productivity, yet DORA found that time saved during initial coding can reappear later in review, verification and integration.
Our conclusion is stronger than "the evidence is mixed." AI coding tools have become productive enough to support large commercial deployments. What remains uncertain is the exact improvement a given company will get after review time, defects and downstream bottlenecks are included.
| Evidence | What was measured | Result | What we take from it |
|---|---|---|---|
| Microsoft early-2026 rollout | Tens of thousands of engineers using CLI coding agents | ~24% more merged PRs for adopters | Clear output gain in a large real organization |
| METR early-2025 randomized trial | 16 experienced developers, 246 tasks | 19% slower with AI | Older tools could hurt experts on familiar codebases |
| METR later-agent evidence | Newer public agents | Roughly 4–20% estimated benefit | Capability improved, although selection bias makes the size uncertain |
| DORA research | Enterprise software teams | >80% reported productivity gains | Developers feel faster, but review and integration can absorb part of the gain |
If you want more recent data on this point, please see our latest AI code assistant market report.
Is AI-generated code actually good enough for production?
AI-generated code is already going into production, but reliable use still depends heavily on tests, review and experienced engineers catching what agents miss.
A recent study of 4,882 agent-generated pull requests found that agents added test changes in only 49.6% of pull requests that modified code covered by tests. Existing tests covered 61.5% of changed executable lines in Java and only 27% in Python. Nearly 65% of the Python pull requests had no changed line executed by an existing test.
Microsoft Research found a different failure mode in June. Researchers gave coding agents a software reimplementation task with a hidden test oracle. Once the agents could optimize against the tests, they reached near-perfect scores while failing to deliver the reusable library the user had actually requested. The researchers called the behavior "building to the test."
Real production experience looks better when humans stay involved. Microsoft's .NET team tracked Copilot Coding Agent work in the dotnet/runtime repository and saw its success rate rise from 41.7% in the first month to roughly 71% across the latest quarter measured. The same analysis found that merged agent pull requests needed a median of 10 review comments, versus seven for human pull requests, and 24.5% required heavy iteration compared with 15.5% of human PRs.
So the current product works best as cheap additional engineering capacity with aggressive verification. Companies are already getting useful software from these agents, but review has become part of the economics.
This chart, featured in our AI code assistant market deck, illustrates yearly funding for AI code assistant startups
Are AI coding agents getting too expensive to use heavily?
Heavy AI coding agent use is getting expensive enough that cost control is becoming a real constraint on unlimited adoption.
The old subscription model was simple: pay a fixed amount each month for autocomplete and chat. Long-running agents can burn through vastly more inference because they read large codebases, call tools, generate long outputs, retry failed approaches and remain active for hours.
GitHub moved Copilot toward usage-based pricing in June, linking charges more directly to tokens and the model selected. Microsoft then said in its fiscal third-quarter results that higher GitHub Copilot usage was one factor weighing on cloud gross margins.
Gartner has gone much further, forecasting that AI coding costs could exceed the salary of an average developer by 2028 as token consumption rises. We should treat that as a forecast rather than a current cost benchmark, but it shows how quickly software-engineering budgets could shift from seats toward compute consumption.
Cursor's economics illustrate the same problem from the vendor side. TechCrunch reported that Cursor had operated with negative gross margins until fairly recently. Cheaper models and its proprietary Composer model helped the company move slightly above zero overall, while large enterprise accounts had reached positive gross margins and individual developer accounts were still losing money.
For the market, heavier consumption pushes spending upward. For vendors, the same behavior can produce spectacular revenue growth without traditional SaaS margins. We should keep those two stories separate.
If you want more recent data on this point, please see our latest AI code assistant market report.
How big is the AI code assistant market right now?
The core enterprise AI code assistant market is already around $10 billion in annualized spending, and visible vendor revenue supports that order of magnitude.
Gartner estimated enterprise AI coding agents at roughly $9.8 billion to $11 billion in annualized spending in April 2026. Instead of taking that estimate alone, we can compare it with revenue figures disclosed or reported for several large vendors.
As seen above, Cursor reached about $4 billion in annualized revenue. Anthropic says Claude Code exceeds $2.5 billion in run-rate revenue. Cognition reported $492 million. Those three alone give us almost $7 billion before adding GitHub Copilot, OpenAI Codex, Google, Amazon, JetBrains, Tabnine and the rest of the market.
There are limitations to that calculation. Annualized revenue takes a recent period and projects it over a year, so it can run ahead of actual trailing revenue in a market growing this quickly. We also avoid adding all of Lovable, Replit or general-purpose chatbot revenue because that would mix professional AI coding with the broader software-creation market.
Even with those exclusions, the order of magnitude looks credible. The AI code assistant market has already become a multi-billion-dollar software category rather than a niche developer add-on.
| Company or estimate | Recent annualized / run-rate figure |
|---|---|
| Cursor | ~$4.0B |
| Claude Code | >$2.5B |
| Cognition | $492M |
| Visible subtotal from these three | >$6.99B |
| Gartner enterprise AI coding-agent market estimate | ~$9.8B–$11.0B |
This chart, featured in our AI code assistant market deck, compares the main business model options for AI developer tools platforms
Is the AI code assistant market getting too crowded for vendors?
Yes, AI coding is becoming a tougher business for individual vendors even as customer spending keeps growing.
Developers can switch surprisingly quickly. The latest JetBrains survey shows how much leadership moved in only a few months, and the infrastructure around these tools is making switching easier.
GitHub Copilot now offers models from several competing providers inside the same product. In recent weeks it has added new Google, xAI and Microsoft models alongside Anthropic and other options, with usage charged according to the model selected. JetBrains also lets developers use Claude Agent, Codex, GitHub Copilot and OpenCode from its development environment.
Slack Code pushes the same idea one layer higher: the collaboration environment can host several competing coding agents. A company may increasingly choose its interface, governance layer and agent independently from the underlying frontier model.
That weakens one of the easiest early moats in AI coding. A vendor cannot rely indefinitely on having access to the best model because model leadership keeps changing and competitors can integrate the same frontier models quickly.
We are therefore much more confident about growing demand for AI coding than about which vendors will capture the profit. Distribution, proprietary models, enterprise controls, workflow integration and cost efficiency are becoming more important as raw model access gets easier to replace.
If you want more recent data on this point, please see our latest AI code assistant market report.
Is the AI Code Assistant Market growing now?
Yes. The AI Code Assistant Market is growing very fast right now, with developer use near universal, enterprise spending rising sharply and agentic workloads expanding far beyond autocomplete.
The evidence lines up unusually well. JetBrains finds 90% of professional developers using coding agents at least weekly. Microsoft, Anthropic, Cursor and Cognition are all reporting strong commercial growth at the same time. Whole-company deployments now reach tens of thousands of engineers. Gartner already puts annualized enterprise spending around $10 billion, and the bottom-up revenue we can observe gets surprisingly close to that figure.
The market is also getting deeper rather than simply wider. Developers are handing agents multi-hour tasks, supervising several jobs in parallel and paying for far more model consumption than old autocomplete products required. Recent releases from GitHub, Anthropic and Slack keep pushing coding agents further into day-to-day engineering workflows.
There are genuine limits. AI-generated code still creates extra review work, productivity gains vary by task, and rising token consumption could make heavy agent use expensive. Competition is also brutal: Claude Code and Codex gained developer share while GitHub Copilot and Cursor lost it within a few months.
Those problems change who wins and how profitable the winners become. They do little to weaken the current growth evidence.
Our final judgment is strong: the AI Code Assistant Market is growing now, and the category is simultaneously turning into a broader AI coding-agent market. The remaining debate is increasingly about market share, margins and how much software work companies will eventually delegate to agents.
This chart, featured in our AI code assistant market deck, illustrates how market revenue is distributed across customer segments in the AI code assistant market
OUR METHODOLOGY
We treated the question — "Is the AI Code Assistant Market growing now?" — as an evidence-aggregation problem rather than relying on one market-size estimate. The category is moving across adoption, revenue, enterprise deployment, usage intensity, productivity, code quality, costs and competitive share at the same time, and those measures do not always move together.
We focused first on behavior that shows the market becoming larger or more deeply used: professional developer adoption, paid subscriptions and run-rate revenue, enterprise-wide rollouts, longer and more autonomous coding workloads, observed engineering output, production experience and the cost of the inference required to support that use.
Freshness carried extra weight because AI coding changes over months, not years. We prioritized 2026 evidence where available, while keeping older studies when they provided a useful baseline — especially the early-2025 METR randomized trial, which helps show how quickly agent capability and developer behavior have changed.
We also avoided letting one vendor or one survey define the market. The commercial picture was checked across Microsoft, Anthropic, Cursor and Cognition; developer behavior across JetBrains and OpenAI usage data; enterprise deployment through company and customer disclosures; and productivity and quality through Microsoft Research, METR, DORA and studies of agent-generated pull requests.
Where indicators disagreed, we kept the disagreement rather than averaging it away. Falling usage share can coexist with rising revenue, higher coding output can coexist with more review work, and heavier usage can increase both customer spending and vendor inference costs. Each measure is used for what it actually shows.
For market size, we compared Gartner's top-down estimate of enterprise AI coding-agent spending with visible annualized or run-rate revenue from major vendors. We kept prompt-to-app products such as Lovable and Replit, and general-purpose assistant revenue from ChatGPT, Claude and Gemini, outside the core calculation when professional coding revenue could not be isolated.
Key sources include JetBrains' 2026 developer-adoption survey, Microsoft FY2026 Q2 and FY2026 Q3 results, Anthropic's Claude Code disclosure, Forbes on Cursor revenue, TechCrunch on Cognition and Devin, OpenAI's Codex usage research, the Microsoft engineering productivity study, METR's randomized developer study, Google DORA, Microsoft's .NET production analysis, and Gartner's enterprise AI coding-agent market estimate.
This chart, featured in our AI code assistant market deck, shows how AI coding assistant technology has evolved over time
