Token Billing, AI Cost Creep.
GitHub Copilot switches to token billing June 1. Here's what that means for accounting firms running AI tools on client workflows.
GitHub Copilot Moves to Token Billing June 1 — Your AI Tool Costs Are No Longer Predictable
Starting today, Microsoft's GitHub Copilot drops flat-rate subscriptions and charges by token usage — meaning the more your team uses it, the more you pay, with no ceiling. This is the first major AI productivity tool to make this switch, and it won't be the last. For accounting firms that have baked AI tool costs into fixed-fee engagements, a usage-based model introduces margin risk that flat subscriptions never did. If your team is running Copilot-assisted scripts against large general ledgers or month-end data pulls, your cost per engagement just became variable.
Audit every AI tool subscription your firm holds — if any are moving to usage-based pricing, your fixed-fee engagements need a cost buffer or a cap.The Accountants Who Understand AI Billing Models Will Own the Client Advisory Conversation
Your clients are running into the same token-billing shock on their own AI tools — Copilot, ChatGPT Teams, Gemini for Workspace — and most of them have no one internally who can explain why costs tripled last month. This is the same moment cloud computing created when AWS moved from reserved instances to on-demand: the people who understood the billing model became indispensable advisors. You already own the cost management relationship with your clients. Add AI spend to your monthly review — it takes one conversation to become the person they call when the bill looks wrong.
Pull one client's AI tool invoices this week and ask them to walk you through what they're actually using — you'll find billable advisory work waiting there.Replit's Agentic Workflows Are Running Real Finance Operations — Here's What That Looks Like in Practice
At SaaStr AI 2026, the Replit team detailed how their own agents — including one called QBee — are handling operational workflows that used to require human coordination across multiple systems. The relevant part for finance leaders: Replit's CEO described 'programming in English' as now genuinely functional, meaning non-engineers can build and deploy automation pipelines without writing code. For AP automation and financial close workflows, this closes the gap between 'we need IT to build this' and 'our controller built this herself over a weekend.' The barrier to automating your own firm's back office just dropped significantly.
If you've been waiting for IT resources to automate your close process, that excuse has an expiration date — English-language agent builders are production-ready now.SoftBank's $75B AI Infrastructure Bet Means Commodity Intelligence Is Coming — And Hourly Billing Is the Casualty
SoftBank is putting 75 billion euros into AI data center capacity in France alone. That level of infrastructure investment means AI inference costs are heading toward zero faster than most firm owners are planning for. When intelligence is cheap enough to be a utility, the billable hour for routine analysis — variance explanations, flux commentary, ratio interpretation — becomes as defensible as billing by the page for photocopies. The firms rewriting their service model around judgment, relationships, and decisions — not the production of financial analysis — are the ones that survive the pricing collapse. One example already visible: Big 4 firms are quietly repricing audit support services downward while pushing advisory retainers up.
If more than 40% of your revenue comes from producing analysis rather than interpreting it, you're in the path of the price compression, not above it.