Today in AI — 26 July 2026

Digest

AI’s centre of gravity today is moving away from model demos and into the messier question of fit: who asked for this, where does it live, and what happens when it acts on our behalf? The pattern is clear: capability is no longer enough; builders now have to design for consent, cost, social norms and failure.

Consent, taste and the consumer boundary

The sharpest signal may be people turning up to libraries to learn how to avoid AI. That is a product warning: if AI arrives as a default rather than a choice, users may treat the feature itself as the problem.

• Librarians are hosting viral ‘Avoiding AI’ workshops for people who are fed up with Big Tech — TechCrunch
• Lorde says AI glasses are ‘not sexy’ — TechCrunch
• Anthropic’s newest ad is creeping people out — TechCrunch

AI leaves the chat box

OpenAI’s hardware moves point towards a fight over the physical interface for AI: keypad, speaker, wearable, phone assistant. The engineering lesson is simple: the interface changes the trust model, because a device in the room or on the desk feels different from a tab in the browser.

• I tried out OpenAI’s new AI keypad — which will be fun for some coders and slightly mystifying to everyone else — TechCrunch
• OpenAI’s first hardware device is reportedly a screenless speaker that can move — TechCrunch
• Apple opens its new Siri AI to everyone with the iOS 27 public beta — TechCrunch

Agents need guardrails, not vibes

The file-deletion warnings around OpenAI’s flagship model cut straight to the product risk in agentic software. If AI can act inside real workflows, permissions, undo, logs and recovery paths become part of the core feature set.

• OpenAI’s new flagship model deletes files on its own, people keep warning — TechCrunch
• Superhuman’s new auto-draft feature almost makes me like AI replies — TechCrunch
• Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer — TechCrunch

The economics are getting visible

AI’s costs are showing up in labour decisions, power grids, token budgets and capital markets. This is the less glamorous layer of the boom: intelligence has inputs, and those inputs need management.

• Monday.com is the latest tech company to blame AI for layoffs — here are 20 others — TechCrunch
• One fallen power line exposed a growing AI data center problem. Here’s how to fix it. — TechCrunch
• DeepSeek reportedly in talks to raise $1.5B, then IPO — TechCrunch
• Anthropic releases new model, Opus 5 — Axios

Culture becomes an interface

Dating and music are sensitive test beds because they expose whether users want AI near personal taste and emotional choice. Good products here will need more than accuracy; they will need restraint.

• The founder of Hinge raised $18M to build a new AI dating service, Overtone — TechCrunch
• Spotify expands its AI push with a ChatGPT-like music assistant — TechCrunch

The takeaway for builders: the next AI advantage may come less from making systems more capable, and more from making them easier to refuse, limit, audit and live with.

---