Count Your AI Subscriptions
Seriously, pull up your credit card statement and count them.
If you’re a developer or knowledge worker who’s been adding AI to your workflow over the past two years, the number might surprise you. ChatGPT Plus or Claude Pro: $20/month. GitHub Copilot: $10. Your IDE’s AI assistant: another $10. Notion AI (now bundled into the Business tier, which bumped the price accordingly). Your Git client’s AI commit suggestions, baked into a paid plan. A writing assistant. A meeting summarizer.
Each of these is pulling from the same handful of underlying models — OpenAI’s GPT, Anthropic’s Claude, Google’s Gemini. But instead of paying the API providers directly at token rates, you’re paying every product team that has wrapped those APIs in a UI and called it a feature upgrade.
That’s the AI subscription stacking problem, and it’s starting to generate real friction.
What BYOK Actually Is
Before going further: BYOK (Bring Your Own Key) and buying your own AI subscription aren’t the same thing, and conflating them is where a lot of this discussion gets muddy.
When you pay $20/month for Claude Pro or ChatGPT Plus, you’re buying access to a consumer chat interface. When you get a developer API key from Anthropic or OpenAI, you’re buying raw model access billed by the token — typically orders of magnitude cheaper for typical usage patterns. A few dozen API calls might cost you under a dollar; the same interaction volume through a consumer subscription costs the same flat fee regardless.
BYOK, in the context of SaaS products, means a product lets you plug in that API key so your requests route through your own account rather than the vendor’s managed infrastructure. The product does the UX and integration work; you pay the AI provider directly.
Who’s Doing What
The landscape is split.
Products offering BYOK: JetBrains launched BYOK for their entire IDE suite in late 2025, letting developers connect Anthropic, OpenAI, or Gemini-compatible keys directly — no JetBrains AI subscription required. Warp, the terminal application, supports BYOK across multiple providers. Factory.ai, Cursor, and Continue.dev treat it as a first-class integration path. These companies are betting that flexible pricing builds more trust than margin capture.
Products sticking with managed AI: Notion bundled unlimited AI into their Business tier in May 2025 and raised the plan price — no BYOK option. GitKraken’s AI features are packaged into paid tiers. Linear has integrated AI into its workflow without an external key path. These aren’t wrong choices, but they’re choices customers are paying attention to.
The split reveals something about the underlying bet each company is making about what their customers value and how AI will evolve as a feature versus a service.
Three Kinds of Users (and What They Actually Want)
The BYOK debate gets noisy because it treats “users” as one blob. They’re not. In practice, most products have at least three segments with genuinely different needs:
Operators want a tool that works. One invoice, one support path, no API keys to manage. Managed AI is almost always right for them. They’re not thinking about token costs; they’re thinking about whether the feature does what it says.
Optimizers track costs, benchmark models, and dislike hidden markups. They already have API accounts. They know how much Claude Sonnet costs per million tokens, and they resent paying a $10/month flat fee for a feature they could run for $0.80. BYOK is right for them, and not offering it feels like a choice made at their expense.
Governed teams need procurement approval, policy controls, and clear data handling boundaries. They may want managed AI for simplicity while also needing enterprise BYOK options so they can route sensitive data under their own account agreements. For them, the answer isn’t one or the other — it’s both, with clear controls over which mode applies where.
Forcing a single monetization model when you serve all three segments is where products create unnecessary churn.
The Genuine Case for Managed AI
Managed AI isn’t just margin capture. It’s legitimate product work.
When a company controls the AI backend, they can lock behavior to tested model versions so a workflow that worked last quarter still works next quarter. They can add fallbacks when a provider has an outage, rather than letting your workflow simply stop. They can update prompts and guardrails centrally when a model release changes behavior. And when something breaks, you file a support ticket — you’re not debugging across your own API account, the product, and a provider status page simultaneously.
For non-technical users and teams, managed AI removes a category of complexity that they shouldn’t have to think about. Cost predictability is also real: finance teams that approve flat per-seat SaaS fees don’t want variable token bills that spike when someone runs a large batch operation.
These are legitimate reasons managed AI earns its price — especially for Operators who value reliability over control.
The Genuine Case for BYOK
And yet the frustration is also legitimate.
The cost math is hard to argue. If you send 50 requests to Claude Sonnet in a month for an IDE feature, your API bill might be under $1. The same volume routed through a $10/month managed integration costs… $10. That gap is hard to justify when you already understand token pricing.
Model choice matters more as the ecosystem fragments. Managed AI locks you to whatever the product team has integrated and maintained. BYOK lets you pick the model best suited for a given task — a cheaper model for simple autocomplete, a stronger one for complex reasoning, a local model for sensitive code. As the model landscape evolves faster than any single product team can track, flexibility has value.
Privacy and data routing become clearer. With BYOK, you know exactly which provider account receives your prompts and under which account-level terms. With managed AI, you’re trusting that the product team has configured their pipeline responsibly. That trust is usually warranted — but it’s trust, not verification.
The “commodity test” is worth running. Ask: if AI inference became a commodity at near-zero margin tomorrow, would customers still pay for your product? If yes, BYOK is not a threat — it’s a trust accelerator. If no, bundled AI revenue may be covering a differentiation gap that will surface anyway.
A Simple Decision Framework
No single model is right for everyone, but the decision isn’t arbitrary either.
Lean toward managed AI when you’re onboarding non-technical users, you need predictable per-seat pricing for finance approvals, or reliability and support accountability outweigh cost control.
Lean toward BYOK when your users are developers who already manage API accounts, cost transparency matters more than convenience, or you need to route data under your own account terms for compliance reasons.
Offer both when you serve multiple segments. The best products increasingly ship managed AI as the default with a BYOK option for users who want it — JetBrains is moving this direction, Warp was there from the start. The tools that will feel out of step are those offering only expensive managed options to users who are already paying the underlying provider directly.
The Market Signal in JetBrains’ Pivot
When JetBrains announced BYOK, the developer community response was telling. It wasn’t “cool, another option.” It was more like a collective exhale — relief that a major vendor had listened and responded.
That signal matters. JetBrains cannibalizing some of its own AI subscription revenue to offer BYOK means the customer frustration was real enough to price into the roadmap. That’s not a minor feature decision; it’s a bet on which customers they want to keep.
Notion made a different bet. Bundling AI into the Business tier at a higher price point means customers who wanted Notion for its core workspace features — but weren’t particularly interested in AI — now either pay more or reconsider. There’s been friction about that choice in the communities where this gets discussed.
Neither company made an irrational decision. They made different bets about what their customers want long-term.
What to Do Right Now
If you’re feeling subscription fatigue, a 20-minute audit is worth it:
- List every AI-powered tool you’re paying for — including tools that have “added AI” to existing subscriptions.
- Identify which ones route through OpenAI, Anthropic, or Gemini under the hood.
- Check whether any offer a BYOK option — you may be surprised how many quietly do.
- For the ones that don’t, decide whether the managed experience is worth the premium given what you’d pay at API rates.
The answer isn’t always BYOK. For some tools and some workflows, the managed experience is genuinely worth what you pay. But it should always be a conscious choice — not a default you accepted without noticing.
If every tool you use added a BYOK option tomorrow, would you flip the switch — or has the managed simplicity earned its place in your stack?