Optimus Consulting
The Week in AI
Edition 09
UK / Service SMEs
This week, the useful AI was the AI no one was watching.
The big launches this week had one thing in common. They let AI run without a person sitting over it. Scheduled tasks, recorded workflows, always-on service agents. AI is moving out of the chat window and into the background of everyday work. That is where the time savings are, and where the new risks hide.
Top stories
5 to knowChatGPT now does the work while you are away.
OpenAI rolled out scheduled Tasks for paid ChatGPT users and a Codex Record and Replay feature that captures a workflow once and reruns it on demand. The prompt box is turning into something that runs jobs on a timer, not just when you ask.
Optimus angle
Pick one recurring Monday task. A report pull, a chase list, an inbox scan. Set it to run on a schedule this week so it is done before you sit down. Then check the first few runs yourself before you trust the output.
Microsoft's Copilot Cowork goes live, and you pick the engine.
Microsoft made Copilot Cowork generally available with consumption-style billing on top of the Microsoft 365 Copilot subscription. Customers can choose between Anthropic and OpenAI models, with hosted DeepSeek reportedly being explored. The question is shifting from which model to pricing, routing and control.
Optimus angle
Being able to switch engines is useful, but a meter rewards discipline. Decide which jobs are worth metered AI, set a monthly ceiling before a busy week sets one for you, and tell the team which model is the sanctioned default.
Salesforce buys Fin to staff every channel with agents.
Salesforce agreed to acquire AI customer service firm Fin for about 3.6 billion dollars, adding an agent that works across chat, email, WhatsApp, text, phone and Slack. Customer service is becoming the first place most firms meet an AI agent running on its own.
Optimus angle
Map the questions your customers actually ask before you let an agent answer them. The repetitive, rules-based ones are safe to hand over first. The judgement calls, and anything touching a complaint or a claim, stay with a person.
After the model ban, firms started building their own backups.
With Anthropic's newest models still under US restriction this week, the practical conversation turned to resilience. Guides on running smaller models locally and keeping a copy of your own data spread quickly. The point was not to self-host everything. It was to not be helpless if access drops.
Optimus angle
You do not need to run your own AI. You do need your work in your own hands. Export your important AI outputs and prompts on a regular basis, and keep one manual path you have actually tested for the day a tool is unavailable.
Every big platform now wants to be your work hub.
OpenAI, Microsoft, Google, Meta and Anthropic all pushed this week to become the place where work happens, bundling agents, memory, file access and even ads into the apps people already use. OpenAI also opened ChatGPT to self-serve advertisers, turning the assistant into a marketing channel as well as a tool.
Optimus angle
Before you buy another AI subscription, list what the tools you already pay for have quietly switched on. Most firms are sitting on capability they have not turned on, and paying twice for the gap.
From the Insurance Desk
Sector signal for brokers, motor claims and credit hire companies.
Item 01
AI could end the quiet insurance renewal.
An Insurance Post editor argued that AI will reshape renewal behaviour, with more than 70 percent of consumers expecting AI to influence their purchases within a year. The automatic, no-questions renewal that brokers rely on is the most exposed.
Optimus angle
When a customer asks an assistant whether to switch, the broker with a clear, accurate online answer keeps the business. Check what an AI assistant actually says about your firm and your products now, well before the renewal run.
Item 02
The AI skills gap is now driving insurance deals.
A new survey found UK insurance chiefs pursuing mergers and acquisitions to buy in AI capability, while warning that skills gaps are widening as adoption lags. The talent to use AI well is becoming as scarce as the tools themselves.
Optimus angle
You do not need to acquire a firm to close the gap. Pick the two roles where AI would save the most hours, train those people properly on one workflow, and you will out-run competitors still waiting for a big platform decision.
Editor's column
The Optimus Take
Last week the AI story was about accountability. Who answers when the machine gets it wrong. This week the story moved on. The big releases were all about letting AI run without someone watching it. A task that fires on a schedule. A workflow recorded once and replayed. A service agent that answers a customer while the office is dark.
That is where the real productivity sits, and it is also where the quiet mistakes happen. The firms that win with background AI will not be the ones who switch on the most. They will be the ones who are clear about what AI is allowed to do unsupervised, and who check the first runs before they look away. Supervised once, trusted slowly, is the rule.
- StrategyDecide which tasks AI is allowed to run unsupervised, and write it down before the meter starts.
- OperationsSchedule one repeatable job this week. Watch the first two runs closely before you trust the third.
- ServiceMake sure customers can reach a human on anything that involves a complaint, a claim or a judgement call.
“An assistant you prompt is a tool. An assistant that runs on a timer is a process.”