The enterprise versus developer divide
The AI coding tools market is experiencing a bifurcation that no one in procurement was prepared for. On one side, enterprise IT departments want a single, auditable, cost-predictable tool that runs through existing SSO, generates compliance logs, and shows up as one line item on the software licence invoice. On the other side, developers are voting with their behaviour: they want whatever produces the best code the fastest, regardless of what the purchasing policy says.
The result is a shadow-stack problem that is playing out at companies of every size. A developer's official tool might be GitHub Copilot. Their actual coding workflow involves Claude Code for multi-file reasoning, Cursor for in-editor autocomplete, and occasionally a locally running model for sensitive codebases. Enterprises are now reckoning with the cost, compliance, and cultural implications of that reality.
According to the latest coding agent leaderboard data, Claude remains the model of choice for complex agentic tasks, but the tooling layer on top of it is fragmenting rapidly. As of June 2026, Cursor, Claude Code, and Codex are converging into what The New Stack describes as "an overlapping stack" — each tool has carved out a niche, but the lines between them are blurring every quarter.
The Microsoft move — what is actually happening
The decision originated inside Microsoft's Experiences and Devices division — the organisation responsible for Windows, Office, and Surface. Engineers in that group received internal communications instructing them to transition away from Claude Code and onto GitHub Copilot CLI before 30 June 2026. Most internal Claude Code licences for the division are being cancelled.
The reasoning is not hard to reconstruct. Microsoft paid Anthropic for Claude Code licences at a time when GitHub Copilot did not have a comparable CLI experience. That gap has since narrowed. More importantly, Microsoft owns GitHub Copilot. Every Claude Code licence fee is a direct payment to Anthropic — a company in which Microsoft has no stake (unlike OpenAI, in which it has invested heavily). From a portfolio logic standpoint, standardising on Copilot makes obvious financial sense.
The timing is also notable. GitHub switched Copilot from request-based to usage-based AI Credits billing on 1 June 2026, giving Microsoft a much cleaner way to track and control developer AI spend internally. Under the new model, Enterprise licences include 3,900 AI Credits per user per month — enough for substantial coding assistance — with a temporary promotional allowance running at three times that figure through 1 September 2026. For a company with Microsoft's engineering headcount, consolidating onto a tool it controls and that now has predictable, scalable billing is a straightforward call.
What the move does not mean is that Microsoft is pulling back from AI-assisted development. Copilot is being actively developed; the billing pivot to AI Credits is itself a signal that Microsoft expects usage to keep rising. The question for engineers inside the division is whether the Copilot CLI can match the agentic capabilities that made Claude Code their preferred tool.
Developers who have built complex multi-session agentic workflows on Claude Code will find the transition non-trivial. The Claude Code multi-session orchestration architecture has no direct equivalent in Copilot CLI as of June 2026. Teams should audit their Claude Code usage patterns before assuming the switch is a like-for-like replacement.
Uber's cautionary tale — when AI tools eat your budget
If Microsoft's decision looks like deliberate portfolio management, Uber's situation looks more like an unplanned collision between fast adoption and slow finance.
In December 2025, Uber rolled out Claude Code and Cursor to its engineering organisation. The rollout was fast and broad — the kind of enthusiastic adoption that makes for good internal AI-strategy presentations. By April 2026, the entire 2026 AI coding tools budget was gone. Four months into the fiscal year, Uber had spent what finance had projected for twelve.
The mechanics of how this happens are not mysterious. Large engineering organisations — Uber has thousands of engineers — consuming per-seat SaaS licences plus per-token API usage simultaneously can generate spend at a pace that quarterly budget models simply do not capture. Cursor's pricing scales with users; Claude Code's usage scales with the complexity and frequency of tasks. When both scale simultaneously across a large organisation that is actively encouraged to use the tools, the spend curve is exponential rather than linear.
The episode reveals a gap in how enterprises are planning for AI tool adoption. Traditional software licence models had predictable per-seat costs. AI coding tools layer per-token consumption on top of per-seat fees, and usage intensity varies enormously across an engineering team — a senior engineer building complex agentic pipelines may consume fifty times the token budget of a junior engineer using autocomplete. That distribution is not something most finance teams have modelled yet.
Uber's situation will not be unique. As more enterprises roll out Claude Code and Cursor to large engineering teams, budget surprises of this magnitude will become a recurring story throughout 2026. The companies that avoid them will be those that instrument their AI tool spend with the same rigour they apply to cloud infrastructure costs — per-team budgets, usage dashboards, and graduated rollout rather than organisation-wide enablement on day one.
The real adoption data — what developers actually prefer
The enterprise pushback against Claude Code is happening against a backdrop of overwhelming developer preference for it. A JetBrains survey of professionals with ten or more years of experience found that 46% named Claude Code as their preferred AI coding tool. GitHub Copilot, despite its massive installed base and corporate backing, came in at just 9%.
That gap — 46% versus 9% among the most experienced developers — is the number that should make enterprise IT departments uncomfortable. It suggests that the developers most likely to be working on the hardest problems, and most likely to drive ROI from AI tooling, are systematically choosing a tool that many enterprises are trying to standardise away from.
Cursor's trajectory tells its own story
Cursor (built by Anysphere, which raised at a $2B valuation) doubled its revenue every two months between June 2025 and February 2026. That is a growth rate that typically only happens when a product has achieved genuine product-market fit — not just trial adoption, but sticky, habitual use by developers who are generating real results.
The revenue growth maps onto what developers report: Cursor's in-editor experience, particularly its Composer mode for multi-file edits, has become genuinely difficult to replicate with alternatives. Teams that started with Cursor as a "nice to have" autocomplete tool found themselves restructuring their entire development workflow around it.
OpenCode enters at the top
June 2026 also brought a new entrant worth watching: OpenCode, an open-source CLI coding agent, launched and immediately hit number one on developer charts. Its significance is less about any single feature and more about what it signals: the open-source community is now building competitive alternatives to every proprietary coding agent, and they are doing it quickly. For enterprises concerned about vendor lock-in or data privacy, OpenCode represents a credible path to AI-assisted development without a commercial dependency.
The convergence that The New Stack identified — Cursor, Claude Code, and Codex overlapping into a single broad category — is now being extended by open-source alternatives. The market that looked like a two-horse race between Copilot and Claude Code six months ago now has five or six serious contenders.
GitHub Copilot's billing pivot — more expensive or smarter?
On 1 June 2026, GitHub moved Copilot from its flat-rate request-based model to a usage-based AI Credits system. The change is significant enough that it deserves careful analysis rather than a reflexive reaction.
Under the new model, credits are allocated per licence tier and consumed based on the model and feature used:
| Tier | Monthly credits | Price/user/month | Promo credits (until 1 Sept) |
|---|---|---|---|
| Pro | 1,500 | $10 | 4,500 |
| Business | 1,900 | $19 | 5,700 |
| Enterprise | 3,900 | $39 | 11,700 |
The reaction from developers tracked by Visual Studio Magazine was pointed: "You will get less, but pay the same price." That characterisation is most accurate for developers who were previously hitting no usage limits under the flat-rate model. For teams that were light users, the credits may represent more value than they were previously extracting.
The more nuanced read is that the billing shift aligns Copilot's cost structure with its actual consumption — the same logic that cloud providers applied to compute and storage. The AI Credits billing switch rewards teams that are intentional about when and how they invoke AI assistance, and penalises teams that treat it as a free-floating background service.
Until 1 September 2026, GitHub's promotional allowance triples the credit allocation across all tiers. Teams evaluating whether Copilot's new billing model works for them have a genuine window to instrument their actual usage patterns before the credits normalise. Use this period to build internal dashboards — the spend surprise that hit Uber can be avoided with three months of good telemetry.
The deeper question the billing change raises is whether Copilot, even at its best, can close the preference gap revealed by the JetBrains data. Credits can be optimised; product quality is harder to adjust on a quarterly roadmap. The developer sentiment problem — that experienced engineers significantly prefer Claude Code — is not one that a billing restructure resolves.
It is also worth noting the context in which Anthropic is competing. Claude Fable 5 launched on 9 June 2026, with Claude Opus 4.8 featuring Dynamic Workflows having launched on 28 May. Anthropic filed a confidential S-1 on 1 June 2026 at a reported $965B post-money valuation. The company behind Claude Code is not standing still.
What this means for AI builders in India and the UK
The enterprise-versus-developer tension described above has a specific implication for builders in India and the UK: multi-tool fluency is becoming a premium skill, not a niche one.
When a Microsoft or an Uber mandates GitHub Copilot but their best engineers are running Claude Code and Cursor in parallel, the person who can navigate both worlds — who can deliver results through whichever tool the enterprise policy allows, while knowing when to advocate for a better tool and how to demonstrate the ROI — is the person enterprises need on the team.
Indian engineering teams are particularly well-positioned here. The combination of strong systems-programming depth, early adoption of AI tooling, and cost-competitive consulting rates means that AI builders who can demonstrate production Claude Code and Cursor workflows — not just familiarity, but shipped projects — are commanding a significant premium in both domestic and international markets.
UK builders face the same dynamic, compounded by the regulatory dimension: as the UK's Frontier AI Bill progresses and enterprises face increasing pressure to document AI tool usage for compliance purposes, builders who understand how to structure AI-assisted development within a governance framework are disproportionately valuable.
The production Claude Code multi-agent orchestration patterns that were considered advanced six months ago are now table stakes for senior AI engineering roles. The builders who have shipped those patterns in real codebases — who have the GitHub commits, the case studies, the production metrics — are exactly the profiles that enterprises are searching for, whether their official policy says Claude Code or Copilot.
Builders with this stack are precisely what enterprises are hiring for
Whether it is Microsoft rationalising its tooling or Uber rebuilding its AI spend model, the common thread is that every enterprise in this story needs developers who genuinely understand Claude Code, Cursor, and Copilot — not just in theory, but in production. If you have shipped with these tools, a Verified Builder profile on AI Tech Connect puts you in front of the hiring managers who are actively looking.
Add your profile free →The market for multi-tool AI coding fluency is, if anything, accelerating. Google's Gemini CLI is expected to be replaced by Antigravity CLI — another signal that the tooling landscape is in active flux and that builders who can adapt quickly across new tools as they emerge will command a lasting premium over those who have only learned a single workflow.
For developers in Bengaluru, Hyderabad, Chennai, London, Manchester, and Edinburgh who are reading this and thinking about their positioning: the window to establish multi-tool credibility before the market matures is open now, not twelve months from now. Browse the Verified Builder profiles to see how the builders already on the platform are framing their AI tool expertise, and consider what your own production work with these tools looks like from the outside.