The gap between who benefits from AI and who gets left behind is widening fast. This edition we look at why smaller companies are falling behind larger enterprises in the AI deployment race, why people trust AI less the more they use it, and whether the industry building this technology actually has a plan for the workers it is displacing.
The trend: AI deployment is growing across organizations, but company size is emerging as one of the clearest dividing lines between those scaling AI and those still running pilots.
The details: Harvard's 2026 AI index report found that organizations with over $5 billion in revenue are nearly 3x more likely to have AI fully scaled than those with under $100 million in revenue. Larger companies also move faster, with more resources that deployment requires. The gap is especially pronounced in agentic AI, where deployment remains in the single figures even at large enterprises, suggesting the challenge compounds further down the size curve.
Why it matters: Productivity and competitive gains from AI come from deployment, not experimentation. If smaller organizations can't bridge the game from pilot to production, the returns will keep concentrating at the top, widening an already significant advantage.
The trend: UK consumers are using AI more than ever, but two problems are emerging. People don't trust the institutions managing their AI data, and users lack enough understanding of the underlying technology to feel totally in control of what they're using.
The details:EY's 2026 AI Sentiment Index found that 74% of people (UK based) had used AI in the past 6 months, but only 43% trust companies to manage their AI data responsibly, and just 41% trust the government to do the same. Institutional skepticism is one thing, but the second problem is that only 14% of people would be comfortable relying on fully autonomous, agent-led systems. Notably, this skepticism was highest amongst those who use AI the most, suggesting that the more people see of AI, the more doubts they have.
Why it matters: While organizations are racing to deploy agents wherever they can, people don't trust them. The technology and those its meant to be helping are heading in opposite directions, and that gap will only get harder to close the more companies deploy agents, without building consumer trust first.
Anthropic and OpenAI launched competing joint ventures on the same day, both built around the same idea: embed AI engineers directly into client operations and get paid like consultants. Anthropic's venture, backed by the likes of Blackstone and Goldman Sachs, at a $1.5 billion valuation, will place engineering teams inside mid-size businesses to rebuild workflows from the ground up. OpenAI's equivalent, The Development Company, is operating at a larger scale with $4 billion raised from 19 investors. The target market is vast. For every dollar companies spend on software, they spend six on services. Sequoia partner Julien Bek argued in April that the next great company won't sell software at all but outcomes: work delivered by AI and billed like consulting.
A new NYT piece by Jasmine Sun captures something that rarely makes it into the public AI discourse: most people inside the industry privately expect the median worker to lose economic leverage as AI automates away their jobs, and they are deeply uncomfortable talking about it on the record. Dario Amodei has warned publicly that 50% of entry-level white-collar jobs could disappear by 2030. AI benchmarks are now explicitly designed to measure how well models replace investment bankers, consultants, and lawyers, with OpenAI reporting an 80% win rate against human professionals. Early-career workers are already feeling it, with employment for software developers aged 22 to 25 down nearly 20% since 2022. A theorized "permanent underclass" is not an inevitability, it is a policy choice, and right now nobody with the power to make that choice is moving fast enough to matter.
Biological data centers could be an energy efficient alternative to silicon-based systems, but are there valid ethical concerns around biological computing?
%201.svg)
