AI This Week: Agents Everywhere, Bills Rising
This week, three forces converged on the same pressure point: the cost and complexity of running AI at your business just changed. A new always-on agent arrived, a familiar billing model disappeared, and a major AI law was rewritten days before it was supposed to take effect.
Google Launches Gemini Spark — A 24/7 AI Agent for $99/Month
At Google I/O last week, Google announced Gemini Spark, an always-on AI agent that runs on dedicated Google Cloud infrastructure even when your computer is off. Unlike the standard Gemini chatbot, which stops the moment you close a browser tab, Spark keeps working: it can monitor your Gmail, pull data from connected apps, draft follow-ups, and queue tasks for your approval — overnight, on weekends, whenever.
The pricing lands at $99.99 per month through the Google AI Ultra subscription. That buys you a capable orchestrator across Google Workspace plus any tool that exposes a standard MCP (Model Context Protocol) endpoint, which means Spark can connect to non-Google tools as well. It is currently available to Google AI Ultra subscribers in the US; Workspace preview for business customers is rolling out shortly.
The practical pitch for a small business: a sales team could configure Spark to scan the CRM each morning for deals with no activity in the past week, draft personalized follow-up emails in Docs, and surface them for human approval before anyone has had their first cup of coffee. That kind of workflow previously required a developer or an expensive enterprise platform.
What this means for your business: If your team already lives in Google Workspace, Gemini Spark is worth a pilot before the Workspace business preview opens to everyone. The $100/month price point is manageable if it reclaims even a few hours of follow-up work per week. Identify one repetitive process — lead follow-up, weekly reporting, inbox triage — and test it there first. Read more at the Google Cloud blog.
GitHub Copilot Switches to Usage-Based Billing Starting Today
As of June 1, 2026, GitHub Copilot is ending its flat-rate subscription model and moving to GitHub AI Credits — a consumption-based system where one credit equals $0.01, and each session consumes tokens billed at the underlying model’s API rate.
The old model was simple: pay a fixed amount per month, use Copilot as much as you want. The new model rewards light users and penalizes heavy ones. Agentic coding sessions — where Copilot is given a task and executes multiple steps autonomously — have been reported to consume $30 to $40 per session. For a developer using Copilot aggressively, that adds up fast.
This is part of a broader industry shift. Microsoft is raising M365 prices in July. Anthropic has moved to tiered usage pricing. The pattern across the industry is the same: the flat-rate era for AI tools is ending, and vendors are moving to billing that scales with how hard you actually push the model. The Goldman Sachs research team flagged this week that AI agents can increase token demand by up to 24 times compared to standard chatbot interactions — meaning agentic billing will almost always cost more than people expect.
What this means for your business: If any of your team members use GitHub Copilot, today is the day to pull your usage data and model what the new billing actually costs you. Set a monthly credit budget per developer now — before the first bill arrives. More broadly, audit every AI subscription you own: which ones are shifting to consumption pricing, and which ones still offer flat rates worth locking in. The GitHub community discussion has specifics on the new credit structure.
Colorado Rewrote Its AI Law Four Days Before Enforcement
In a significant policy reversal, Colorado Governor Jared Polis signed SB 26-189 on May 14 — a replacement for the original Colorado AI Act that was days away from its June 30 enforcement date. A federal magistrate judge had already stayed enforcement of the original law on April 27.
The original law was expansive: it required businesses that deploy “high-risk” AI systems — those influencing decisions in hiring, healthcare, housing, lending, or education — to run formal risk management programs, conduct annual impact assessments, and actively mitigate algorithmic discrimination. The replacement bill strips most of that back to a narrower notice-and-transparency framework. The new law takes effect January 1, 2027, and enforcement depends on the attorney general completing a rulemaking process that hasn’t formally started.
The short version: the strict version of the Colorado AI Act is gone. The replacement is lighter, later, and still being defined. That said, the Colorado situation is not isolated — 1,200 active state-level AI bills remain in play nationally, and the EU AI Act is still on track for full enforcement on August 2, 2026. Businesses with any European customer exposure need to confirm software vendor compliance before that date.
What this means for your business: If you were preparing for Colorado’s June 30 AI compliance deadline, that deadline is no longer live — but don’t cancel the work entirely. The replacement law still requires transparency notices for AI-driven decisions, takes effect in seven months, and the EU August deadline is real. Use the extra time to build a simple AI vendor log: what each tool does, what data it touches, and whether your vendor has signed a data processing agreement with you. That log is the foundation for any compliance response, under any law.
The Real Cost of AI Is Getting Harder to Ignore
This week’s data points from Microsoft and Goldman Sachs landed at the same place: AI is more expensive to run than most businesses budgeted for. Fortune reported that Microsoft’s own internal research is surfacing an uncomfortable finding — using agentic AI at scale can cost more than the human employees it is meant to replace. Uber’s CTO disclosed that the company burned through its entire 2026 AI coding tools budget in four months.
None of this means AI is a bad investment. It means that the “pilot it and the savings will be obvious” phase of AI adoption is ending. Businesses that deployed AI tools without measuring outcomes are now getting their first real billing statements from consumption-based vendors — and the numbers are not always what the sales decks promised.
The companies holding up best are the ones that treated AI deployment the same way they treat any other operational expense: with baselines, targets, and accountability. For a 10-person business, that does not require a formal ROI framework — it requires a spreadsheet and a monthly check-in.
What this means for your business: Before adding any new AI subscription, establish what you expect it to do and how you’ll measure whether it’s doing it. Time saved, leads contacted, drafts reviewed — pick one metric per tool and track it for 90 days. If the number doesn’t hold up after usage-based billing kicks in, cancel early. See Josh Bersin’s breakdown of the enterprise pricing shift for more context on where this is heading.
The Takeaway
The story connecting this week’s news is accountability. The free-trial era of AI is closing. Flat subscriptions are becoming consumption bills, AI laws are moving from drafts to enforcement calendars, and early adopters are learning what their AI tools actually cost when the honeymoon ends. The businesses that come out ahead are the ones measuring what they get — not just what they spend. If you want a second set of eyes on your current AI stack and what it’s actually delivering, reach out to us.