More than 500 communities are blocking data centers. AI has a trust problem.

Welcome back.

The most important AI stories this week were not really about smarter models.

They were about everything required to make those models useful.

Communities are resisting the infrastructure that powers AI. xAI has packaged agents so they can operate the same software as your team. And one AI company is hiring someone whose entire job is to help employees redesign how they work.

The common thread is simple: the model is only one part of the system.

Power, permissions, processes, skills, and public trust are becoming just as consequential.

Let’s get into it.

Before we get into it, I have a quick favour to ask. We are learning how readers use AI Ready and what would make it more useful. Share your feedback in this short survey.

AI is running into the real world

The number of active local restrictions on new data center developments in the United States reportedly rose from roughly 300 in late June to more than 500 in July.

The resistance crosses political lines. Communities are raising concerns about electricity prices, water use, noise, pollution, and how much public infrastructure these facilities consume. New York has paused approval for facilities using 50 megawatts or more, while Texas has also slowed new grid connections as it studies their impact.

This will not make your AI tools disappear tomorrow. But it challenges one of the assumptions behind the AI boom: that companies can continue adding computing capacity whenever demand increases.

For founders, model cost and availability should not be treated as permanent background conditions. They depend on physical infrastructure, regulation, energy, and public support.

That matters when a workflow only produces value because inference is unusually cheap or a single provider is absorbing the real cost.

Build the workflow around a measurable outcome. Then ask whether the economics still work if prices rise, limits change, or you need to switch providers.

AI may feel like software. Its constraints increasingly look like heavy industry.

Grok Bot gets its own computer

On Tuesday, xAI introduced Grok Bot, a team of agents that can sign into workplace tools, operate their interfaces, and continue working when you step away.

Each bot uses a computer in the cloud. You can show it how a process works, save that process as a routine, and allow several bots to coordinate on the same job. xAI says its own teams have used them for sales research, CRM updates, invoices, recruiting, and bug fixes.

The technology underneath this is not entirely new. Computer use, persistent memory, and agent coordination have existed separately for some time.

What has changed is the packaging.

Instead of asking someone to configure a complicated automation, xAI wants them to message a bot as if they were delegating to a colleague.

That makes agents more accessible. It also makes their mistakes more consequential.

A bot that can use every tool does not need unlimited freedom. It needs a narrow responsibility, clear permissions, approval points, and a definition of success.

Start with one bounded job. Give it the minimum access required. Measure whether it completes the work reliably. Expand only after it earns more responsibility.

Calling it a digital employee does not make it accountable. The operating system around it does.

The next important AI hire may be a teacher

Basis, an AI company building agents for accounting firms, is hiring an AI Education Lead.

This is not a conventional training role. Basis wants someone who tests new models, creates playbooks for different roles, helps employees redesign their own work, and measures whether people are actually working differently.

One line in the description captures the distinction: the person owns adoption, not attendance.

That is a much better standard than the one most companies use.

Many businesses buy AI licenses, run a workshop, count active users, and call the rollout successful. But using a tool is not the same as improving a business.

The better questions are:

• Did execution become faster?

• Did quality improve?

• Did the cost of the process fall?

• Can someone else repeat the result?

A smaller company may not need a dedicated AI Education Lead. But someone still needs to own this work.

Their job should not be to promote AI. It should be to identify worthwhile workflows, help teams redesign them, establish review rules, and prove whether the change produced value.

Without that owner, AI adoption usually remains a collection of individual experiments.

ONE CONVERSATION WORTH FOLLOWING

Josh Miller recently asked why most people outside technology still do not seem to care about agents, even as the products become dramatically more capable.

My read is that this is not simply an awareness problem.

Most people do not want an agent. They want a frustrating part of their job to disappear without creating a new management burden.

The agent products that break through will not be the ones with the most impressive demos. They will be the ones who reliably finish recognizable work.

IN OTHER NEWS

• Grok 4.6 arrived with a greater focus on long-running agent work. xAI says it matches GPT 5.6 Sol on the Artificial Analysis Intelligence Index, with pricing starting at $2 per million input tokens and $6 per million output tokens.

• OpenAI slowed parts of its Astra release after testing suggested the model may have reached a critical level of cyber capability. The uncomfortable question is whether safety testing can keep pace with the systems it is evaluating.

• Google reorganized DeepMind’s leadership, moving Demis Hassabis toward science and long-term strategy while placing operational responsibility elsewhere. Even frontier labs are discovering that research leadership and execution leadership are different jobs.

That’s all for this week. See you on Tuesday.

Haroon

P.S. If this helped you make sense of the week, forward it to the person responsible for AI inside your company. If that person does not exist yet, that may be the first problem to solve.