Notion Rebuilds AI Pricing Around Credits, Not Seats

Notion just ripped up its AI pricing model in favor of a credit system with model routing and bring-your-own-AI options, Mistral shipped a voice-native agent model that skips the text-to-speech pipeline entirely, and a wave of tracked people (Karri Saarinen, Dan Shipper, Shreya Shankar) are independently circling the same question: as agents get cheap and capable, what's actually left for a product to own. ### Notion - Agents, Rebuilt to Do More for Less: A New Credit-Based Pricing Model **Source:** https://www.notion.com/blog/notion-ai-more-for-less **Credibility:** High (first-party product announcement from co-founder Akshay Kothari) **What happened:** Notion rebuilt its AI product around three changes: smarter model routing (so cheaper models handle simpler requests automatically), long-running agents that can work for extended periods without check-ins, and a new credit-based consumption system that replaces whatever pricing structure existed before. Notion also added a bring-your-own-AI option, letting users or orgs plug in their own model access instead of routing everything through Notion's backend. **Key technical details:** - Model routing automatically assigns requests to different models based on task complexity, instead of running everything through one fixed model - Long-running agents can execute multi-step work without constant user prompting (consistent with Notion's recent custom agent workflows for internal teams, covered previously) - New credit system replaces the prior AI consumption/pricing structure, which Notion frames as "more for less" - Bring-your-own-AI lets customers use their own model provider relationship rather than exclusively Notion-brokered access **Why it matters for PMs:** This is a pricing architecture decision, not a feature launch, and it's the more interesting part. Moving from a flat or seat-based AI allowance to a credit system with routing is Notion betting that usage will become spiky and unpredictable as agents run longer and do more autonomous work. The bring-your-own-AI option is the tell: Notion is decoupling "using AI in Notion" from "paying Notion for the AI itself," which only makes sense if they expect model costs to keep compressing and don't want pricing friction to be the thing holding back agent adoption. If you're pricing your own AI features on a per-seat or flat-rate basis, this is a preview of where that model breaks down once agents start running unsupervised for longer stretches. **Critical questions:** - What happens to a customer's experience when they run out of credits mid-task on a long-running agent? Does work get interrupted, queued, or degraded to a cheaper model? - Does bring-your-own-AI cannibalize Notion's own AI margin, or is it aimed at enterprise customers who already have model contracts elsewhere and would otherwise not adopt Notion AI at all? - How transparent is the model routing to end users? If Notion silently downgrades a request to a cheaper model, does quality become inconsistent in ways users notice but can't diagnose? **Action you could take today:** If your product has any AI feature priced on a flat or seat basis, model what happens to your margins if usage patterns shift from "a few prompts per session" to "an agent running unattended for an hour." Notion just told you this shift is coming.