Optimus Consulting
The Week in AI
Edition 06
UK / Service SMEs
The cost of AI moved from flat fee to meter.
AI got more capable and more metered in the same week. Copilot switched to usage billing, agents went always-on, and investors poured billions into the firms behind the tools. The new question for operators is not what AI can do. It is what it costs to run, and whether that cost can be predicted.
Top stories
5 to knowGitHub switches Copilot to pay-by-usage. Developers revolt.
GitHub moved Copilot to token-based billing, charging by how much you use rather than a flat seat fee. Many developers reacted badly, worried that costs are now harder to predict from month to month. It is a small change with a big signal. The AI tools inside your daily software are moving to meters.
Optimus angle
Find every AI tool your team pays for and put it on one page. Note which bill by seat and which bill by usage. Set a monthly cap on the usage-based ones before a busy week sets it for you.
Anthropic ships Claude Opus 4.8 and raises 65 billion dollars.
Anthropic released Claude Opus 4.8, a stronger model for reasoning and agentic work, and confirmed a 65 billion dollar raise at a valuation nearing one trillion ahead of a possible IPO. The capability you rent keeps improving. So does the capital riding on you renting it.
Optimus angle
Capability is rising faster than most firms can absorb. Do not chase every model upgrade. Pick the one or two tasks where a better model saves real hours, and leave the rest until they do.
The token paradox. Cheaper per task, higher per month.
Several reports flagged the same pattern. The price of a single AI request keeps falling, yet total AI spend keeps rising, because agentic tools do far more work per job. A lower unit cost is hiding a higher monthly bill for teams that do not track usage.
Optimus angle
Measure AI spend the way you measure any variable cost. Per seat, per task, per month. A tool that looks cheap per request can still be your fastest-growing line if nobody is watching the total.
Google's Gemini Spark runs as a 24/7 agent.
Google rolled out Gemini Spark, an always-on assistant that works in the background around the clock, alongside a pricier Gemini 3.5 Flash. Early reviews call it useful. Always-on is the point, and also the catch. An agent that never sleeps is an agent that never stops metering.
Optimus angle
Before you set an agent running around the clock, decide what it is allowed to touch and what it must hand back to a person. An always-on tool needs an always-on boundary.
Gartner warns 40 percent of agentic projects could be scrapped by 2027.
Gartner forecast that 40 percent of agentic AI projects may be cancelled by 2027, often over weak controls and unclear returns. The same week, buyers across the market sounded more selective about AI spend. The mood is shifting from build anything to prove it pays.
Optimus angle
Give every AI pilot a number to hit before you start. Hours saved, errors cut, cost removed. A pilot without a target is the first one a board cancels.
From the Insurance Desk
Sector signal for brokers, motor claims and credit hire companies.
Item 01
The FCA tests insurer AI in the open.
Avantia became the sole insurer to share its experience inside the FCA's AI Live Testing initiative, where the regulator assesses real-world AI use directly. For brokers and claims firms, this signals that AI in regulated workflows will be examined, not assumed safe.
Optimus angle
If you use AI anywhere near advice, pricing or claims decisions, write down what it does and where a human signs off. The firms ready for FCA questions are the ones who can show the workflow, not just the tool.
Item 02
A quarter of insurance buyers are firmly against AI.
New research found over a quarter of UK insurance customers are strongly opposed to AI in the sector. As brokers and credit hire firms add AI to claims and service, customer trust is now a live commercial issue, not a side note.
Optimus angle
Be clear with customers about where AI helps and where a person stays in charge. For credit hire and claims, a plain line on who handles their case can protect trust while you automate the admin behind it.
Editor's column
The Optimus Take
Six editions in, the pattern is clear. AI capability arrives first, and the bill arrives second. This week the bill caught up. Copilot moved to usage pricing, Gemini went always-on, and even the firms building these tools are raising tens of billions to keep the lights on. None of that is free, and increasingly none of it is flat.
For UK service SMEs the lesson is not to spend less on AI. It is to spend on purpose. A tool that saves ten hours a week can carry a rising bill. A tool nobody can point to a result for cannot. The difference is whether you are measuring.
- StrategyDecide which AI spend is buying a real business result, and cut anything that cannot point to hours saved, errors reduced or revenue moved.
- OperationsPut every AI subscription on one page with cost, billing model and owner. Set a monthly cap on every usage-based tool before a busy week sets it for you.
- ServiceGive one person the job of watching the AI bill each month. Rising usage without a rising result is the first thing they should flag.
“Cheaper per token is not cheaper per month.”