What GitHub built matters more than the headline price change. They introduced a custom intermediary currency, the AI credit. That sits between the customer and the underlying model costs. Seat pricing stays, but a monthly credit pool now layers on top, burning down based on whichever models, features, and workflows the team uses.

A custom currency that abstracts the unit of value from the unit of cost isn't new. Canva does it with Magic Credits for generative design, Midjourney builds its pricing around GPU minutes, and Salesforce ships Agentforce on a per-conversation unit. Each arrived at the same conclusion independently: when underlying infrastructure costs are variable and growing, a custom currency gives the provider a controllable abstraction layer and gives the customer a stabilized, governable unit they can plan around.

The custom currency model does something else that is easy to overlook: it gives the vendor the ability to price against business outcomes rather than infrastructure units. A “conversation” in Salesforce is not a token count, but a proxy for a customer interaction. A GitHub AI credit is not a raw model cost by another name, but an abstraction that can be repriced as GitHub’s model mix, feature set, and margin targets evolve. **The currency decouples the pricing surface from the infrastructure reality underneath it.**

There's an obvious objection here: token costs have plummeted since 2023. If the underlying cost keeps dropping, shouldn't flat-rate pricing get more sustainable, not less? Why introduce a custom currency at all?

Token prices for non-frontier models are still trending down, but two forces most organizations underestimate are offsetting that decline.

First, context depth is growing faster than token prices are falling. Average sequence lengths in production LLM workloads have exploded, driven by agentic inference, tool calling, and longer conversation histories. A single MCP tool call can spawn dozens of sub-calls, and subtle prompt changes can radically shift token consumption, often invisibly to the consumer. Context windows have expanded from 4K to 8K tokens in 2023 to 128K to 1M today, and filling more of that window costs more regardless of per-token rate improvements. Cheaper tokens don't help if tokens per operation are climbing faster.

Second, and more strategically, a credit currency isn't just a cost pass-through. It's a pricing surface that can map to any unit of value the business chooses: compute time, API requests, agent actions, data processed, or outcomes delivered are all equally valid. Organizations that lock pricing to raw tokens are betting tokens stay the right proxy for value indefinitely. A custom currency keeps that question open. As AI capabilities evolve and the link between model cost and business value gets less direct, that abstraction layer is exactly what preserves pricing flexibility.

*AI workloads and agentic workflows carry higher costs of goods sold (COGS) compared to traditional solutions. If unaccounted for, these become a margin-bleeding liability for solution providers. The orgs building AI products today with no metering infrastructure are, in effect, running at an unknown cost structure. The custom currency model is more than a pricing strategy: it is the financial foundation for treating AI like a business rather than a science project.*

A reasonable objection: GitHub, Canva, and Salesforce adopted credit currencies because they had massive market share first. Millions of users gave them leverage to change pricing terms. A mid-market enterprise or a developer building on the same infrastructure doesn't have that leverage. Is the credit model accessible to anyone? Or is it a mega-platform privilege?

The market share argument is real, but it conflates two separate problems: the power to enforce a pricing change on an existing user base, and the ability to launch with a credit model from the start. 

GitHub's transition drew loud detractors because it moved existing customers from one contract to another. An organization building an AI product today faces no such constraint. Starting usage-first, before there's a user base to upset or contracts to renegotiate, is far easier than evolving into it later under margin pressure or feature subsidy fatigue.

The companies that will face painful transitions in 2–3 years are the ones launching AI products today on flat-rate or per-seat models, assuming they'll figure out monetization later. 

"Do we have enough users to change pricing?" Wrong question. Ask yourself: Do you want to build this migration problem into your roadmap, or avoid it entirely?"

GitHub’s rollout is a cautionary tale for any organization considering a similar move on an existing product. The community response was significant. Power users — the ones who had built real productivity around Copilot’s agentic features — felt taxed without anything new in return. Meanwhile, lower-maturity users with idle seats saw an upside. Same product. Same price change. Totally opposite experiences depending on how much value you were actually extracting. 

The change in monetization model shifted the market fit alignment and perception of value.

This market-fit shift wasn't accidental. It wasn't done on a whim. Changing how you charge people creates tension. That's just the reality of any contract shift between a provider and the community it serves.

The organizations that handle this well don't just hand heavy users a higher bill. They make the case based on clear value. A credit pool with real governance — budget ceilings, cost-center allocation, user-level visibility, overage policies set in advance — makes consumption feel like control, not punishment.

The technical infrastructure for metering is the solvable part. The harder work is the governance design: who sets pools, who approves overages, how teams are allocated capacity based on need and demonstrated value. Integrating governance like that from the start marks the distinction between a managed transition with positive adoption and the chaotic crisis of a reactionary response.

The credit model is now a demonstrated pattern across the enterprise software market. GitHub, Canva, Salesforce, Midjourney, and the LLM API providers themselves have all converged on it. Will this become the standard pricing architecture for AI products? It already is. Will your organization build for it from the start, or manage the migration later?

For platform teams, ISVs, and enterprises building AI products on top of infrastructure they control, that means three concrete requirements: metering every AI workload at the runtime layer before costs spiral; defining a credit model that maps usage to business outcomes rather than raw token counts; and enforcing entitlements at the point of access — not in a reporting dashboard after the fact.

GitHub’s pricing move further signals the winds of change in economic realities of service economics. Seat-based or flat-rate access to AI capabilities and agentic workloads is not a viable pricing model when underlying costs are variable, context depth is growing, and the relationship between model usage and business value is increasingly non-linear. It appears that every platform that has tried to hold flat-rate pricing under serious agentic usage has eventually arrived at the realization that custom currency models aligned consumption-based metering is non-optional. The timing varies. The destination does not.

The organizations that operate from a position of strength in the AI economy are the ones that treat metering and monetization as first-class infrastructure concerns, not afterthoughts. That means building the monetization models into the delivery pipeline before there is a user base to upset, before there is pricing debt to manage, and before margin pressure forces a rushed migration. It means defining what your custom currency represents in terms of business outcomes and not just what it costs in the COGS of tokens, calls, and compute cycles. And it means giving your users, customers, internal teams, agents, the governance tools to manage their consumption before the bill arrives.

The custom currency economy is not a GitHub story. It is the pricing architecture of the agentic AI era. The builders who start with a value-unit first approach will not have to explain to their best customers why their economics just changed.

*Watch the on-demand webinar *[*Monetizing the Agentic Era*](https://konghq.com/events/webinars/monetizing-the-agentic-era)*Monetizing the Agentic Era** and get a technical blueprint for building a scalable AI monetization system that turns token consumption into revenue.*