Welcome back to Good Better Best.
Each week, we break down real pricing, packaging, and product moves from SaaS and AI leaders and share the ideas worth stealing.
This week, we’re digging into a pattern we’ve observed around agent credit pricing: the broader the job, the more granular the pricing unit. The narrower the job, the easier to price based on a higher-level action, output, or outcome. We’ll dig into a recent update by n8n to show how this looks in practice.
Before we get there, a quick look at some of the most impactful moves of the last week. Let’s get to it.
🔌 PricingSaaS Partners power the next era of SaaS + AI Monetization
This Week in Pricing, Packaging, and Product
This week we observed 200+ changes. The highlights:
Klaviyo swapped its mobile/SMS credit allowance for a flat $5/mo [Link]
Getty Images raised UltraPack credit pack prices [Link]
Freshworks removed its free starter program [Link]
HubSpot increased the email template limit on Sales and Service Hub [Link]
Buffer removed its "Audience demographics" feature [Link]
Groq dropped Llama 4 Scout and Qwen3 32B [Link]
Mighty Networks added MCP support to the Scale plan [Link]
Notion delayed Workers Beta credit billing to Oct 15 [Link]
Shopify swapped flat SMS bundling for pay-per-send pricing [Link]
Stripe disclosed a 100M events/month quota in a Meters API footnote [Link]
Devin upgraded the Pro plan's bundled free model [Link]
Supabase swapped its flat pipeline fee for a metered hourly rate [Link]
Temporal swapped the Enterprise support add-on for a Dedicated Platform Architect [Link]
Thinking Machines Lab added Inkling models and retired Kimi-K2.5 [Link]
Krea narrowed Unlimited generations to relaxed-only [Link]
Synthesia launched a 450+ free dubbing minutes promo [Link]
Together AI added 3 new models to Serverless Inference pricing [Link]
LangChain overhauled LangSmith billing [Link]
WorkOS dropped its top-tier SSO discount [Link]
Check out more updates on PricingSaaS →
The broader an AI agent's job, the more granular the pricing unit becomes.
When an AI feature performs exactly one job, every invocation costs roughly the same. Vendors can charge for the user-facing action—a build, a message, a request—and absorb the small differences in inference cost themselves.
The moment an AI product expands beyond a single job, the cost of each interaction becomes wildly variable. At that point, charging per action either underprices power users or overprices everyone else. The pricing unit naturally shifts downward to the underlying resource: tokens, inference, or compute.
n8n's latest pricing update is a great example of this pattern.
n8n introduced a new AI Assistant to replace its AI Workflow Builder, and adjusted credits accordingly.
It's worth placing this change within the context of n8n's broader pricing strategy.
n8n has always separated running automations from building them.
Workflow Executions meter production usage.
AI Assistant Credits meter AI assistance during development.
That distinction is important because n8n isn't changing how customers pay for automation. They're changing how customers pay for building, maintaining, and evolving automation.
Previously, n8n offered an AI Workflow Builder. It was a purpose-built feature that did one primary job: generate workflows. Every prompt consumed one Builder Credit, regardless of how complex the request was.
Recently, n8n expanded that product into a broader AI Assistant. Instead of simply generating workflows, the assistant now helps users create, edit, debug, explain, troubleshoot, and iterate on existing automations. As the scope of the product expanded, the pricing model evolved alongside it. Builder Credits were replaced with token-based AI Assistant Credits that burn according to inference consumed rather than interactions.
This shift from a workflow generator to an always-on development companion means an "interaction" stops being a meaningful pricing unit. Two conversations with the same number of prompts can consume dramatically different amounts of compute.
As a result, the number of included credits changed dramatically. Starter increased from 50 Builder Credits to 2,300 AI Assistant Credits, while Pro increased from 150 to 13,700 AI Assistant Credits per month. Enterprise no longer lists a fixed allocation, instead showing "Coming Soon."
A few implications stand out with this update.
Comparability disappears. It's now much harder to determine whether the product became more or less expensive. The credits no longer represent the same thing. Old credits measured interactions; new credits measure inference. It’s apples to oranges.
Risk shifts from vendor to customer. Flat per-action pricing meant n8n absorbed fluctuations in AI costs. Token metering protects n8n's margins regardless of how expensive an interaction becomes, while customers assume more variability in how quickly they consume their allocation.
Pricing UX becomes a competitive advantage. The tradeoff for aligning pricing with compute is reduced predictability. Early community feedback has focused on generic "out of credits" messages, limited visibility into remaining usage, and the inability to purchase additional credits without upgrading plans. This is the hidden cost of token pricing. Once pricing becomes opaque, the usage meter, forecasting tools, alerts, and top-up experience become part of the pricing model itself — not just the product experience.
The broader lesson extends well beyond n8n.
Product scope and pricing granularity tend to move together. As AI products evolve from purpose-built features into generalized assistants, pricing naturally migrates away from user-facing actions and toward the underlying compute that powers them. But the inverse is just as true. When a product narrows its focus to a well-defined, predictable job, companies gain the ability to abstract away the underlying complexity and price against a simpler, higher-level unit of customer value (e.g., outcomes).
Thanks for reading! If you’re working on AI monetization and want to learn more about how we help, book time here.
Until next time,
Rob