The vibe coding stack: what to pay for and what to skip

A vendor-neutral analysis of the vibe coding stack identifies only two paid tools worth buying from day one: the AI editor Cursor ($20/month) and the terminal agent Claude Code ($20/month on Claude Pro). The report warns that Cursor's metered pricing can turn a $20 plan into a $40 month, and heavy Claude Code users may need the $100 Max plan due to usage caps. App builders like v0, Bolt, and Lovable are deemed optional, as Stack Overflow and JetBrains surveys show the editor-and-agent core dominates developer adoption.

The Primer · Tools & Stacks The vibe coding stack: what to pay for and what to skip Vendor-neutral Two slots earn the money: the editor and the agent, $40 a month at sticker. Here is the whole vibe coding stack, slot by slot, what each one really costs, what the field runs, and the version that costs nothing. Andrej Karpathy named it in a February 2025 post https://x.com/karpathy/status/1886192184808149383 : a way of building where you “fully give in to the vibes” and let the model write the code. Ten months later Collins made vibe coding its Word of the Year for 2025 https://www.thebookseller.com/news/vibe-coding-is-collins-word-of-the-year-2025 . The tools caught the wave, and the pricing pages got complicated. Here is the whole answer up front. The vibe coding stack is four slots, and only two are worth paying for from day one: an AI editor Cursor, $20 a month and a terminal agent Claude Code, on the $20 Claude Pro plan . That is $40 a month at sticker. Everything else, the app builders, a second assistant, an extra chat subscription, is optional. Every price here comes from the vendor’s own page, cross-checked against our maintained price index /palette/ai-tool-price-index/ . What the field runs Before the slot-by-slot, the map. In Stack Overflow’s 2025 survey https://survey.stackoverflow.co/2025/ai/ , 84% of developers use or plan to use AI tools, up from 76% a year earlier, and the tools they name most are the core slots and their cheap cousin: ChatGPT at 81.7%, GitHub Copilot at 67.9%, and Claude Code at 40.8% https://survey.stackoverflow.co/2025/ai/ . The prompt-to-app builders trail in single digits: v0 at 9.1%, Bolt at 6.5%, Lovable at 5.7%. JetBrains’ January 2026 survey https://blog.jetbrains.com/research/2026/04/which-ai-coding-tools-do-developers-actually-use-at-work/ of over 10,000 developers puts Cursor and Claude Code tied for second at work behind Copilot, and gives Claude Code the highest loyalty scores in the category, a 91% satisfaction rating and an NPS of 54. None of this tells you what to buy. It tells you the editor-and-agent core is where the working population sits, and the builder slot is a minority tool. The two slots that earn the money The editor: Cursor, $20 a month. In-context work: edits, refactors, test scaffolding, the tab-complete that keeps you in flow. The catch is the meter behind the plan. Cursor’s $20 Pro includes about $20 of model usage https://cursor.com/pricing , and once that is spent, continued use is billed as on-demand usage in arrears at API rates. It is opt-in, not a surprise charge, and completions stay unlimited, but hand-picking frontier models is how a $20 plan becomes a $40 month. Cursor learned this in public. In June 2025 it restructured Pro https://cursor.com/blog/june-2025-pricing from a fixed request count to that metered pool, “unlimited” turned out to mean only the cheaper Auto model, users got surprise bills, and Anysphere apologized and issued refunds https://techcrunch.com/2025/07/07/cursor-apologizes-for-unclear-pricing-changes-that-upset-users/ . Set the spend limit on day one. The agent: Claude Code, $20 a month on Claude Pro. The terminal agent is the half of vibe coding the editor cannot do: hand it a whole task and let it run /primer/how-to-use-claude-code/ . Claude Pro includes Claude Code https://claude.com/pricing on one shared usage pool, so a heavy agent afternoon eats the same allowance as your chats. When the agent becomes your main tool, the $100 Max plan, up to $200 for the 20x tier https://support.claude.com/en/articles/11049741 , stops being extravagant. Anthropic added weekly caps in July 2025 https://techcrunch.com/2025/07/28/anthropic-unveils-new-rate-limits-to-curb-claude-code-power-users/ aimed at the heavy tail it estimated at under 5% of subscribers, a figure a later class action disputes, so the ceiling is real if you run the agent hard. This is not a fringe tool. Anthropic reports Claude Code past a $2.5 billion run rate https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation as of February 2026. But does it even make you faster Here is the part the vendor pages will never show you. When METR ran a randomized trial https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/ in early 2025, experienced open-source developers were 19% slower completing real tasks with AI tools, and the tool they mostly used was Cursor Pro. The developers thought they were 20% faster. That gap between felt speed and measured speed is the whole risk of a metered stack: you cannot feel your real velocity any more than you can feel your real spend. The other side is real too. A GitHub-run experiment found 55.8% faster https://arxiv.org/abs/2302.06590 on a fixed coding task, and METR’s own February 2026 update https://metr.org/blog/2026-02-24-uplift-update/ walked the slowdown toward zero on newer tools, though it calls that follow-up data unreliable. The point is not to skip the stack. It is to measure, rather than vibe, whether it pays off, which is the only reason to read a bill at all. The optional slot: the app builder Lovable $25 a month , Bolt $25 , and v0 $30 per user sell the same promise: describe the app, get the app. For a non-coder shipping a first product, the slot can be worth it. If you already run the editor and the agent, you can usually skip it, and the meter is why. Lovable prices by request complexity, a styling tweak around 0.50 credits, adding authentication around 1.20 https://docs.lovable.dev/introduction/credits-and-usage , debugging is billed like any other request with no exemption, and a deployed app keeps drawing credits from the same pool while it runs. Bolt counts tokens, $25 for 10 million a month https://bolt.new/pricing , and burns them faster as a project grows, because it re-syncs your files to the model on every message. None of that is a scandal; it is metered pricing doing its job, the same trap the $20 AI app walks into /study/unit-economics-of-wrapping-an-llm/ . The mistake is reading the subscription as the cost when the meter is the cost. What to skip, and why GitHub Copilot $10 . The cheapest seat and, by both surveys, the most-adopted AI coding tool, so if you want one paid tool and live in VS Code it is a defensible sole pick. But it fills the same slot as Cursor. Its billing also changed on June 1, 2026 https://github.com/features/copilot/plans : premium requests became dollar-denominated AI Credits, $15 a month on Pro, while code completions stay unlimited. Paying for Copilot and Cursor is buying one slot twice. Windsurf. The standalone editor is gone; it is Devin Desktop now. That competes for the agent slot, which is already taken. A second chat subscription. ChatGPT Plus is $20 and Codex rides along with it. If your agent is Claude Code, that is a same-slot duplicate. One frontier subscription used hard beats two used at half attention. The $0 stack A real free stack exists in 2026, and it is better than it used to be. Gemini CLI https://blog.google/technology/developers/introducing-gemini-cli-open-source-ai-agent/ gives 60 model requests a minute and 1,000 a day at no charge with a personal Google account, a genuine free agent rather than a teaser. Pair it with Copilot Free https://github.com/features/copilot/plans , 2,000 completions and 50 chat requests a month with agent mode included, or Cursor’s free Hobby tier, and you can vibe code a real thing for nothing. What breaks first is the caps: 50 Copilot chats or a day’s Gemini quota vanish fast in an agent loop, and Claude Code has no free tier at all. The free stack is for finding which slot you would use every day. Then pay for that one first. The bill | Slot | Tool | Sticker | A heavy month | |---|---|---|---| | Editor | Cursor Pro | $20 | $40 to $50 with frontier models | | Agent | Claude Code Claude Pro | $20 | $100 on Max, if it becomes the main tool | | App builder | optional | $0 | $25 to $30 if you take the slot | Total | $40 | $60 to $150 | The sticker total is $40. A heavy month runs higher, and our estimate is $60 to $70, which is an estimate on purpose: the metering is documented on every plan, but no vendor publishes what an average build spends. That is the reason to check the bill against the price index /palette/ai-tool-price-index/ once a month and cancel the slot that stopped earning. How the tools fit together day to day is its own piece /primer/ai-coding-stack/ ; this one is the money. One email, when there's something worth sending Get the receipts in your inbox. No fixed schedule, no filler. You get an email when we've tested something, run the numbers, or found a tool worth your time. Free. Double opt-in, unsubscribe in one click. Running a different stack? Compare notes in the forum ↗ https://community.okaneland.com Sources | Source | Link | |---|---| | Stack Overflow 2025 Developer Survey, AI section. 84% of developers use or plan to use AI tools up from 76% ; out-of-the-box tool use: ChatGPT 81.7%, GitHub Copilot 67.9%, Claude Code 40.8%, v0 9.1%, Bolt 6.5%, Lovable 5.7%; 60% favorable, 46% distrust accuracy vs 33% trust. | |