Claude Gets a Big Upgrade

Anthropic released Claude Opus 4.8, the newest version of its top-tier model. The headline upgrade is consistency on long jobs. It is better at coding, at running multi-step tasks on its own, and at the kind of professional work that takes more than one prompt to finish.

That last part matters more than it sounds. Most everyday AI tools shine on quick questions but lose the thread on big projects. A model that can stay focused across long, running work is closer to a teammate than a search box.

Anthropic also launched Claude Fable 5, a new model it describes as a 'Mythos-class' model made safe for general use. In plain terms, it took a powerful model and added strong guardrails so most people can use it safely.

But the guardrails sparked pushback. Cybersecurity researchers told TechCrunch that Fable's limits are too strict for legitimate security work, blocking tasks they need to do their jobs. It is a classic tension: tight enough to be safe, loose enough to be useful. Anthropic is leaning hard toward safe.

The most aggressive AI adopters are now spending roughly $7,500 per employee every month on AI tools, according to the Ramp AI Index. That is a striking number, though as TechCrunch notes, it is still less than an engineer's salary.

The takeaway is not that you should match it. It is that a real spending category is forming. AI is moving from a free trial line item to a budgeted cost, the way software and cloud bills did a decade ago. Plan for it before it surprises you.

AI is quietly moving into everyday operations. McDonald's is testing a Google-backed system nicknamed 'Archy' that takes drive-thru orders and helps run restaurants, currently in trials at five locations.

On the defense side, insurer Aviva uncovered a record 230 million pounds in fraud claims and is using AI to fight back. As the report notes, fraudsters now come armed with their own AI tools, so it is increasingly machine against machine.

Datadog veterans launched Niteshift, an AI coding startup, with a $7 million seed round. Their bet is simple: companies will want control over which AI models they use, not to be locked in to one big provider.

This is a smart thing to watch even if you never touch the product. Lock-in is a real risk with any vendor. Building so you can swap models later keeps your costs and your options open as the market shifts.