The April 2026 Stack Overflow developer flash poll, the JetBrains developer ecosystem survey, and the GitHub usage data triangulate to roughly the same number: three products — GitHub Copilot, Cursor, and Claude Code — together hold somewhere between 68% and 74% of the AI coding tools market by active developer usage. The exact number depends on whether you measure by daily active users, monthly active users, or paid seats, but the picture is consistent across measurement methods.
This is unusually high concentration for a developer tools market. For comparison, the IDE market is split across roughly seven products that each hold 5-20% share. The version control market is functionally a one-product market (Git) but the hosting layer is split across three providers. The cloud infrastructure market is split across three providers with significant share each.
AI coding tools have compressed to three-product dominance in about 30 months. The concentration carries real risk for developers, real opportunity for the three winning vendors, and real consequences for the second tier of products that are now competing for the remaining 30% of share.
The Share Picture, Triangulated
The three primary data sources for April 2026:
Stack Overflow flash poll (n=4,200 developers, April 22 2026): - Copilot: 31% - Cursor: 23% - Claude Code: 17% - Codex CLI: 11% - Gemini CLI: 6% - Other / none: 12%
JetBrains 2026 Q1 ecosystem survey (n=12,800, fielded Q1 2026): - Copilot: 34% - Cursor: 19% - Claude Code: 14% - Codex: 9% - Tabnine: 6% - Continue: 4% - Other: 14%
GitHub usage telemetry (Q1 2026 internal, partially publicly disclosed in the April 2026 Developer Velocity report): - Copilot accounts for the majority of AI coding tool usage on GitHub-hosted repos. Specific figures undisclosed but consistent with 30%+ market share when extrapolated.
The averages across the three sources land Copilot at roughly 31-33%, Cursor at roughly 19-23%, Claude Code at roughly 14-19%. Total for the three: 64-75%.
The pattern is consistent: one dominant share leader (Copilot), one fast-growing challenger (Cursor), one premium-positioned third place (Claude Code). The remaining 25-35% is fragmented across Codex, Gemini CLI, Tabnine, Continue, Aider, Windsurf, and several smaller products.
How Each Product Won Its Share
The three winning products got there through different paths.
Copilot won on distribution. GitHub's installed base, the deep integration with VS Code (which itself dominates the editor market), the Microsoft enterprise sales motion, and the price point ($10/month individual, $19/month business) combine to make Copilot the default AI coding tool for any developer who has not actively shopped for an alternative. Capability has been a follower rather than a leader through the product's history, but the distribution advantage is structural.
Cursor won on UX. The editor-resident agent mode, the chat-with-codebase feature, the multi-line edit suggestion flow, and the speed of the iteration loop produced a product that developers prefer using even at a higher price ($20/month) than Copilot. Cursor's share growth from low single digits in early 2024 to nearly 20% in early 2026 is the fastest-rising story in the category.
Claude Code won on per-task capability. The Opus 4.7 underlying model, the parallel task forking, the harness design that extracts peak performance from the model, and the focus on repo-wide edit quality produce measurably better outcomes on hard coding tasks. The share is the lowest of the three but the developer cohort is the most committed — Claude Code users report the highest net promoter score among coding agents.
The three winning paths are not substitutable. Copilot's distribution moat cannot be replicated by smaller vendors. Cursor's UX moat is replicable but takes years of focused product investment. Claude Code's capability moat requires owning the underlying model, which only Anthropic, OpenAI, and Google can do at the frontier.
The Concentration Risk Decomposed
For developers, the 70% concentration carries four distinct risks.
Vendor dependence at the workflow layer: A developer who has spent 6+ months building muscle memory around Cursor's keyboard shortcuts, prompt patterns, and workflow integrations faces real productivity loss if they need to switch tools. The switching cost compounds with usage time, and the three winning products all benefit from this lock-in.
Pricing power as competition recedes: A three-product market gives the leaders meaningful pricing power. Copilot, Cursor, and Claude Code have all held their per-seat pricing flat through the recent capability improvements, but the structural ability to raise prices is now meaningful. If the second tier of products continues to weaken, expect prices to drift up by 20-40% over the next 18 months.
Model-vendor coupling: Each of the three winning products is tightly coupled to specific underlying models. Copilot to OpenAI (primarily GPT-4.x/5.x with some Anthropic options), Cursor cross-vendor but with OpenAI as the default, Claude Code Anthropic-only. A developer's coding tool choice is now also a model vendor choice, which constrains their ability to mix-and-match for different workload types.
Tooling diversity at risk: The second tier of products has historically driven feature innovation in the category. Aider pioneered git-aware coding, Continue pioneered the "Cursor-but-open-source" pattern, Tabnine pioneered local-model inference for privacy-sensitive teams. As these products lose share, the rate of feature innovation outside the top three declines, and the top three set the pace for the category.
Where The Second Tier Still Holds Ground
Concede the obvious: the second tier of coding tools has not been outright defeated. Several products have meaningful share in specific developer segments.
Codex CLI at 9-11% share is growing faster than Claude Code in some segments (notably developers who heavily use the OpenAI ecosystem). The multi-surface architecture (CLI + IDE + Cloud + Inbox) is a structurally different bet than the top three, and the share trajectory is positive.
Gemini CLI at 6-11% is the youngest of the products with double-digit share. The free tier and open-source release in April 2026 may compound through Q2 and Q3. The plausible upside scenario is reaching 15%+ share by year-end if the open-source plugin ecosystem develops.
Tabnine at 4-6% has held a specific niche in privacy-sensitive and air-gapped enterprise deployments. The local-inference capability is unmatched in the top three and represents a structural advantage for the cohort of developers and teams that cannot send code to cloud-hosted models.
Continue at 3-5% has carved a niche as the open-source extensible editor agent. The plugin architecture and the cross-model flexibility position it as a workshop product for developers who want to customise their workflow rather than adopt a pre-built one.
Windsurf at 2-3% has the strongest competing editor-resident agent product but has struggled to differentiate against Cursor's faster product velocity. The share trajectory is roughly flat through Q1 2026.
The second tier is not dead. But the share gap between any of these products and the top three is now wide enough that catching up would require either a structural feature advantage or a major distribution shift.
The Forward Picture
Three scenarios for how the concentration plays out through 2027.
Scenario A — Concentration deepens to 80%+. The top three continue to compound their advantages. Copilot leverages GitHub distribution to extend share. Cursor's product velocity continues. Claude Code's capability lead on the depth segment locks in the premium cohort. The second tier loses ground and the market becomes a three-product oligopoly. This is the trajectory the data currently supports.
Scenario B — Gemini CLI breaks the three-way structure. Google's open-source release plus free tier captures share faster than expected. By 2027 the market is four-product with Gemini CLI at 15-20%. The pricing pressure from a free tier alternative caps the price growth of the paid products. This scenario requires Google to sustain investment in the CLI and the plugin ecosystem to develop on the scale the open-source release implies.
Scenario C — A new product reshapes the market. Either an existing vendor pivots (e.g., JetBrains shipping a first-party agent that beats Cursor's UX, leveraging the JetBrains user base) or a new entrant produces a capability that the top three cannot match (e.g., a truly local-running frontier-tier model). This scenario is the most speculative and the least predicted by current data, but the speed of category change makes it non-zero.
The base case is Scenario A. The most plausible disruptor is Scenario B given Google's commitment to the open-source path. Scenario C is the wildcard.
The Counterfactual For Developers
If the three-product market had not consolidated this quickly — if the share were spread across five products at 10-15% each rather than three at 20-30% — the developer experience would look different. More cross-product feature parity pressure. More aggressive pricing competition. More tooling diversity for specialised workloads. Less switching cost compounded into any single product.
The counterfactual is not necessarily better. A more fragmented market would also mean less product velocity per dollar of vendor investment, less depth in any single product, and harder choices for developers about which tool to commit to. The current state — three strong products with clear differentiation — has real benefits.
For developers in May 2026, the practical implication is to be deliberate about which of the three winning products to invest workflow muscle in, because the switching cost compounds rapidly. Picking the one that matches the workload type (Copilot for default editor-integrated AI, Cursor for premium editor-resident agent, Claude Code for depth coding work) is more important than picking the one with the highest benchmark scores. The capability gap between them is narrow. The workflow lock-in is wide. The decision is one that will shape the next 18 months of how the developer ships code.