The Credit Index reads: the coding tools

If you want to watch AI pricing evolve in fast-forward, watch the coding tools. It is the hottest category in software — the fastest revenue ramps anyone has ever seen, the largest startup acquisition ever announced, a land-grab with real money behind it — and it is also the category that has repriced its meter on nearly everyone who uses it. Those two facts are connected. This is the first in a series reading the AI Credit Index one category at a time; we start here because coding shows, more clearly than anywhere else, what happens to pricing when a market is growing too fast to bother optimizing margins — and what it does the moment it has to.

The hottest category in software#

Start with the velocity, because it sets up everything else. These are, by reported numbers, some of the fastest-scaling businesses in the history of the industry. Cursor went from roughly $100M to a reported $2.6–3B of annualized revenue in not much more than two years. Lovable says it hit $100M faster than any software company before it, and is reportedly around $500M now. Emergent — an India-based builder aimed at non-technical business users — announced $100M in annualized revenue inside about eight months. Devin’s maker, Cognition, reports something close to half a billion. None of these figures are audited; they’re company-disclosed run-rates, and you should read them as such. But the direction is unmistakable, and it’s pointed almost straight up.

The capital agrees. AI startups reportedly took on the order of $212B in venture funding in 2025 — roughly half of all global VC — and coding is consistently named the single largest enterprise use case for generative AI. When a category grows like this, the contest is for share, not for margin. You price to get users in the door and worry about the unit economics later. Hold that thought; it’s the key to the whole pricing story.

And the consolidation has already started. The clearest sign that this is a category worth fighting over is who’s now buying in. Windsurf alone got carved up three ways in mid-2025: OpenAI agreed to buy it for a reported ~$3B, that deal collapsed, Google then acqui-hired its founders and core research team in a reported ~$2.4B licensing arrangement, and Cognition bought what remained — the product, the team, the contracts — for a reported ~$250M, pairing it with Devin. Cognition has since raised at a reported $26B. And the headline of the year: SpaceX agreed in June 2026 to acquire Anysphere, the company behind Cursor, for $60B in an all-stock deal expected to close later in the year — the largest acquisition of a venture-backed startup ever announced, nearly double the $32B Google paid for Wiz that closed in March 2026 (the largest yet completed) — folding the category leader into the same orbit as xAI, which SpaceX absorbed earlier in 2026. (It’s announced, not yet closed, and subject to regulatory approval — but the number is the point.) When the biggest platforms in technology are paying tens of billions to own the coding tools, the people pricing those tools are not, yet, in the business of protecting margin. They’re in the business of not losing.

Two camps, and where they start#

The catalogue looks crowded until you notice it splits cleanly in two, and the split predicts almost everything about how each one prices.

On one side are the pro-developer tools — software for people who already write code. Cursor is the leader here: an AI-native code editor (a VS Code fork) built for professional engineers, starting at $20/month. GitHub Copilot is the incumbent, with the unbeatable distribution of living inside the editor developers already use and a reported ~4.7M paid subscribers, starting at $10. And Cognition now owns both Windsurf (an agentic IDE, from $20) and Devin (billed as “the AI software engineer,” an autonomous agent that takes whole tickets, scaling from $20 to $200).

Two more players belong in this camp but price by entirely different rules: the model labs themselves. OpenAI’s Codex and Anthropic’s Claude Code are full agentic coding tools — terminal agents that take real work — yet neither is sold as a standalone seat. Each is folded into the lab’s flagship subscription (Codex inside ChatGPT, from $20; Claude Code inside Claude’s Pro and Max plans, $20 to $200) and metered against the very same usage window that powers the chatbot, with an optional pay-per-token API path for heavy automation. For a lab, the coding agent isn’t a product to monetize on its own — it’s a reason to buy and keep the model subscription. That’s a different kind of pressure on everyone else here: how do you sell a $20 metered coding seat when the lab hands you its agent inside the $20 you’re already paying it for the model? The labs sit outside the table below for exactly that reason — they don’t price the tool, they price the model.

On the other side are the non-technical app builders — software for people who want a working app without writing the code. Lovable is beloved for turning a prompt into a fast, polished MVP or mockup (from $25). Bolt does in-browser, full-stack prompt-to-app for indie builders (from $20). v0, from Vercel, specializes in design- and prompt-to-UI — front-end generation wired straight into Vercel’s deploy stack (from $30 a seat). Replit is the browser-based, agent-first builder that takes you from natural language to a deployed app with no local setup (from $25). And Emergent aims squarely at mid-market prosumers solving business problems — CRMs, inventory tools, internal ops apps — with a voice-to-app agent and a user base it says is around 70% non-coders (from $20).

Tool	Camp	Starts at	Known for
Cursor	Pro-dev IDE	$20	AI-native editor for engineers
GitHub Copilot	Pro-dev (incumbent)	$10	In-editor default; distribution
Windsurf	Pro-dev IDE	$20	Agentic IDE (Cognition)
Devin	Autonomous agent	$20	”The AI software engineer”
Replit	Browser builder	$25	Natural-language → deployed app
Lovable	Non-technical	$25	Fast MVPs and mockups
Bolt	Non-technical	$20	Indie prompt-to-app
v0	Non-technical	$30/seat	Design/prompt → UI front-end
Emergent	Non-technical	$20	Mid-market business apps; voice-to-app

Notice how compressed the entry prices are: a $10 floor, then a wall of $20–$30. That clustering is not a coincidence — it’s what a land-grab looks like. Nobody wants to be the expensive one at the door. The differentiation, and the money, moved somewhere the pricing page doesn’t advertise: into the meter.

They all marched right#

Read these nine off the metering continuum the Index uses — access-priced on the left, usage-priced on the right — and the first thing you see is that not one of them sells a flat seat anymore. Every tool here meters. They divide into the two rungs nearest the usage end: the non-technical builders (Lovable, Bolt, v0, Emergent) sit at a hard allotment — a fixed pool of credits, and when it’s gone you stop or buy more by hand — while the developer tools (Cursor, Copilot, Windsurf, Devin, Replit) sit one rung further right at metered overage, where you keep working past the allowance and the bill keeps climbing automatically.

And most of them moved there during the window. Cursor went from a soft meter to overage. Copilot went from a flat “unlimited” seat all the way to token-metered overage. Windsurf went from a hard credit pack to overage. v0 went from message limits to a metered dollar allotment. The whole category drifted the same direction, toward charging for what you consume, and the reason is the same one underneath all of it: an AI coding agent burns real, uncapped compute every time it runs. A flat price over that is a margin landmine — the heaviest users, the ones who love the product most, are the ones who blow it up. So the meter wasn’t optional. The only question was how to install it without losing the land grab.

How they repriced — without moving the sticker#

Here’s the part that earns this category its place in the Index. Almost none of them raised the headline price. They changed the meter behind it — the most common move in the whole study, and the coding tools are its purest expression:

• Cursor held its $20 Pro plan and reworked what it covers, so the same money bought roughly half the usage and the old unlimited-slow safety net went away. The backlash was sharp enough to force refunds and a public apology. (Reviewed)
• GitHub Copilot re-based from a flat seat to per-request, then to token-denominated credits at the same headline prices — heavy agentic users reported bills 10–50× what they paid before. (Reviewed)
• Lovable switched from one-credit-per-message to complexity-weighted credits, so a hard task now spends several credits where it used to spend one — same $25, less work. (Reviewed)
• Replit moved to “effort-based” pricing, replacing a flat per-checkpoint charge with a dynamic per-action cost that swings from cents to several dollars. (Reviewed)
• Devin nudged its compute-unit price up, then replaced units with dollar credits and retired its cheapest no-minimum plan. (Reviewed)
• Windsurf went the other way on visibility entirely — it replaced a countable credit balance with opaque daily and weekly quotas. (Reviewed)

The pattern is so consistent it’s almost a category signature: keep the number on the page still, and let the number on the invoice move.

The barbell: cheap to start, expensive to lean on#

Now the split earns its keep, because the two camps repriced differently, and the difference traces straight back to what they do.

The non-technical builders held flatter. Lovable, Bolt, v0, and Emergent mostly stayed at a hard allotment around $20–$30, and some got more generous — Bolt raised its monthly token allowance at an unchanged price. That’s affordable to do because their cost per user is more bounded: someone building an MVP or a business app generates a burst of work and then stops to use the thing they built. The value is real but the consumption is lumpy and finite, so a fixed pool that you occasionally top up is a fair trade for both sides.

The developer tools are where the squeeze lives, because professional engineers run agents all day, every day, and an autonomous agent’s compute cost has no natural ceiling. So Cursor, Copilot, Windsurf, Devin, and Replit all landed on metered overage — and built premium tiers ($100, $200 a month) for the heaviest users on top. The shape is a barbell: a cheap, almost loss-leader entry to win the seat, a metered middle that scales with intensity, and an expensive top for the power users who’d otherwise be unprofitable. You’re cheap to start and expensive to lean on — which is exactly right when what you charge for has to track the cost you can’t control.

Opacity as a shield#

One move deserves its own mention, because it’s where this category went somewhere the rest of the Index mostly hasn’t. When Cognition reworked Windsurf and Devin, it didn’t just change the price of a credit — it removed the credit. Visible, countable balances gave way to daily and weekly “usage” quotas described only in words, with the reassurance that “most users will never hit their limits.” That’s repricing behind frosted glass: if the buyer can’t count the units, the buyer can’t notice when each one starts buying less. It’s the logical end-point of holding the sticker still — first you change the meter quietly, then you make the meter itself impossible to read.

The backlash that changed nothing#

Every one of these repricings made noise. Cursor issued refunds and an apology. Copilot users revolted. Lovable’s complexity-weighting lit up Reddit. v0 users reported $20 allotments vanishing in days. And here is the lesson the whole category teaches at once: none of them walked the model back. v0 added a small daily login credit and kept its meter. Cursor explained itself and kept its meter. The customers didn’t actually leave for a flat-rate competitor — there barely is one anymore — and they didn’t reject usage pricing in principle. What they punished was the manner: the silence, the surprise, the sense of a deal changed underneath them. The companies that handled the change with grandfathering, notice, and a clear reason kept their customers. The ones that sprang it lost trust they had to buy back. How you move a meter matters more than where you move it to — and coding is the category that proved it at scale.

What it means — and what comes next#

For anyone buying these tools, the takeaway is concrete: the entry price tells you almost nothing, and last quarter’s effective price tells you almost nothing either. Translate the meter into your own real work, watch what a credit actually buys this month versus last, and assume it will keep moving — because in this category it has, every few months, for everyone.

And the bigger arc is just beginning. Today’s aggressive, experimental, share-first pricing is a function of a market growing too fast to care about margin. That won’t last. When the growth curve bends — and when the new owners writing $60B checks want a return on them — the pressure flips from winning seats to earning margin, and the meter is the lever that’s already installed. The coding tools repriced once on the way up. The more interesting reprice will be the one on the way to profitability.

Next in the series: we read the Index on a category that priced itself the opposite way.