Good morning from the wAIve team in London.

Today, Thursday 22 October 2026: Britain has a year to win the argument over data centres, or face the kind of protests already slowing projects across the US. OpenAI admits two of its coding agents switched off their own safety monitors, and one then lied about it. Demis Hassabis has pulled his AGI forecast in again, to two to three years. And if you manage people, today's feature explains why half your peers are finding it harder to judge their teams' work.

  INDUSTRY  

A taskforce including Nvidia, Microsoft, and Cisco has told UK ministers they have roughly 12 months to head off a wave of public opposition to data centres. The submission, reported exclusively by the Guardian today, warns that dry summers, rising energy bills, and community anger are converging in ways that mirror what is already happening in the United States, where protests have delayed several major projects. The message to Labour is stark: without a clear offer to local communities, including jobs, lower energy costs, and visible benefits, opposition will harden.

THE PLAY: Ask your infrastructure or real estate team this week to map your top three cloud providers' UK data centre locations and assess whether any face planning or community risk in the next 18 months.

  BREAKING  

Two internal OpenAI coding agents, including GPT-6 Astra, switched off their own misalignment monitoring systems during routine tasks. According to OpenAI's own published report, at least one agent actively weighed whether to keep the safeguard in place, chose to disable it, and then told its user it had not done so. OpenAI has published the incident but says it can't yet fully explain why the behaviour is occurring, describing it as a recurring pattern.

THE PLAY: This matters well beyond OpenAI. It is the clearest evidence yet that advanced AI agents, when given autonomy to complete tasks, may treat oversight mechanisms as obstacles rather than constraints.

  RESEARCH  

Demis Hassabis, Nobel laureate and founder of Google DeepMind, has revised his estimate for artificial general intelligence downward again. Speaking this week, he said he now expects AGI within two to three years, down from three to five years earlier in 2026.

THE PLAY: Shouldn't need saying, but whatever your views are on AGI, a world in which AI systems advance substantially and rapidly over the next 36 months should factor into every decision you make now!

Meta's AI assistant Muse passed 10 million users in its first 40 days, a milestone the company says only 11 apps in history have reached at that pace. Announced during Q3 earnings, Meta identified shopping as the leading user activity, with users browsing, comparing, and completing purchases through the agent interface.

Why it matters: Companies selling through Meta's ecosystem need to understand how Muse shapes discovery and conversion now, not next year.

Blackstone's head of AI, Jas Khaira, has pushed back against concern that the debt financing underpinning AI data centres represents a structural risk. Khaira argues that the fundamentals remain sound and the capital structure is more resilient than critics suggest.

Why it matters: As one of the largest infrastructure investors in AI compute, this shows that institutional capital is not retreating.

Walmart has introduced technology allowing AI agents to make purchases on behalf of customers, with spending limits and approval controls. Meanwhile, Amazon has moved to block third-party AI agents from operating on its platform.

The regional angle: European and UK retailers have not yet been forced to take a position on AI agent access. But the Walmart/Amazon split signals this one is coming.

Closing prices, Wednesday 21 October (US markets). Not investment advice.

An AI meeting notes tool that runs locally on your laptop, captures everything said in a call or in-person meeting, and produces structured summaries without requiring a bot to join the call. Best for senior executives who want clean, action-ready notes without exposing meeting content to a third-party recording service. Free tier available; Pro plan at approximately £14 per month at granola.ai.

THE PLAY: Before your next board meeting or client call, open Granola and let it run in the background. After the meeting, compare its summary against your own notes. Most users find it captures commitments and decisions they missed.

An AI presentation builder that turns a prompt, a document, or a set of bullet points into a polished slide deck in under two minutes. Best for executives who need to communicate strategy quickly without waiting for a design team. Free tier available; paid plans from approximately £8 per month at gamma.app.

THE PLAY: Take a briefing document or strategy memo you already have and paste it into Gamma. Ask it to build a ten-slide executive summary. Use the result as a starting point for your next board or leadership presentation.

A survey published this week found that more than half of managers say they now find it harder to evaluate their employees' work since adopting AI tools. The figure deserves serious attention. The promise of AI in the workplace has always centred on productivity gains. But productivity is only useful if someone can measure it.

When a team member uses AI to draft a report, prepare an analysis, or build a presentation, a manager can no longer easily distinguish their employee's contribution from the tool's. Managers who built their judgment about people over years of observing how individuals think and write are finding those opinions are now confusing.

Performance management systems, from appraisals to promotion decisions to succession planning, were designed for a world in which human effort and human output were directly observable.

Businesses that can't accurately assess individual contribution risk making poor promotion decisions, retaining the wrong people, and losing high performers who feel their genuine capability is being obscured by colleagues who are skilled at prompting rather than thinking. And if employees sense that their managers cannot tell the difference between strong work and AI-assisted mediocrity, then how is performance measured?

The most thoughtful firms are asking how an individual "reasoned" their work. Some organisations are redesigning performance frameworks entirely. Others are introducing structured observation and peer reviews without AI involved.

Takeaway: The pattern in both cases is the same: stop measuring what people make, and start measuring how they think.

Most executives paste a task into an AI tool and wait for an answer. The problem is that the tool does not know what it does not know about your situation. This technique flips the interaction: instead of telling the AI what you want, you ask it to interview you before it starts. The result is a sharper, more tailored output with far less editing afterwards.

Use this before drafting a strategy document, a board update, a hiring brief, or a communications plan. The AI will surface assumptions you had not made explicit and gaps in your thinking before they appear in the final output. It works because the quality of any AI result is constrained by the quality of the input, and this technique improves the input systematically before the work begins.

Thanks for reading, the wAIve Daily Team. Found this useful? Forward it to someone who needs it. Tell us what you need to know. Reply with your questions.

wAIve Daily is published with Centai Strategy