29 May 2026: Claude Opus 4.8 puts agents on notice (AM)

Today’s AI news is less about one dazzling demo and more about the plumbing behind everyday use. Anthropic is pushing Claude towards bigger agent work, Google is packaging more of Gemini for normal users, and infrastructure companies are starting to rebuild around machine traffic rather than human clicks.

Anthropic has released Claude Opus 4.8, with new controls for effort and a research preview of Dynamic Workflows for Claude Code. According to Anthropic’s announcement, Opus 4.8 is available today at the same standard price as the previous Opus release. The company says it improves benchmark performance, gives users more control over how much effort Claude applies to a task, and lets Claude Code coordinate larger pieces of software work through Dynamic Workflows.

The useful point for readers is not the benchmark race itself. It is the shift from asking a chatbot one question to asking a system to plan, call tools, check its own work, and explain where it is uncertain. Anthropic says Dynamic Workflows can split large coding jobs across many parallel subagents, then verify outputs before reporting back. That is a serious claim, so treat it as something to test in narrow, reviewable workflows rather than a licence to let an agent roam through a live business system.

There is also a consumer angle. Anthropic says users on claude.ai can now choose how much effort Claude spends on a response, trading speed and rate limit use against deeper work. That is the kind of control that matters if you use AI for research, writing, or analysis. It also fits the pattern covered in our plain English guide to AI agents: the next wave is not just smarter answers, it is delegated work with more visible controls.

Google has turned its I/O 2026 announcements into a practical map of where Gemini is heading next. The company published a roundup of 12 I/O moments, including Gemini Omni and Gemini 3.5 Flash. The important part is that Google is no longer presenting AI as a separate product lane. It is folding models, search, video, assistants, and developer tools into the same operating rhythm.

For small businesses, the question is whether that integration reduces friction or creates another layer of lock in. If Gemini tools can move from search to writing to video to app actions without constant copying and pasting, they could save time on routine marketing and research work. But the more these tools sit inside one account ecosystem, the more carefully teams need to think about permissions, data access, and who can approve output before it reaches customers. Our guide to using AI to write and edit is still the right baseline: treat AI output as a draft, not a finished judgement.

Sesame has launched a public iOS preview of its conversational AI app, bringing voice first agents closer to normal phone use. TechCrunch reports that the startup, founded by former Oculus leaders and others, has released an iOS app with four named agents and a conversation style designed to feel less like typing into a chatbot. According to Sesame’s own framing, the system tries to balance quick replies with time to search, retrieve, and compose better answers.

The practical test is whether voice AI can be useful without becoming intrusive. Sesame says its agents can use search cards, notes, texting mode, memory, and an incognito mode that saves nothing to memory. Those are the right categories of control, but users should still assume that any voice assistant needs careful privacy settings. The real watch point is whether conversational agents become the front door for everyday tasks before smart glasses and other wearable devices arrive.

AWS and other infrastructure companies are adapting services for AI agents that create short, intense bursts of machine traffic. TechCrunch reports that AWS has launched a new generation of OpenSearch Serverless designed for agentic workloads, with compute that can scale up for bursts and scale back down when idle. Cloudflare also told TechCrunch that bots already account for a significant share of HTTP traffic.

This sounds abstract until a business starts paying for it. Agents can query documents, call APIs, search databases, and run background tasks at a speed that makes old hosting assumptions look expensive. If infrastructure can scale around those bursts, AI agents may become cheaper to deploy. If it cannot, the hidden cost of automation will show up in cloud bills, rate limits, and unreliable workflows. The next useful AI tool may depend as much on boring infrastructure as on model quality.

Asana’s acquisition of StackAI shows workplace software vendors are racing to own the agent layer inside ordinary business systems. TechCrunch reports that Asana has bought StackAI, a no code workflow automation company that builds agents connected to tools such as Salesforce, Slack, and Google Workspace. Asana is positioning the deal around AI teammates and AI Studio, not as a separate experiment.

That matters because many small firms will not build agents from scratch. They will meet them inside the project management, document, and customer systems they already use. The upside is less setup. The risk is that automation starts making decisions across too many connected apps without enough review. The thing to watch over the next few weeks is whether vendors add clearer audit trails and approval controls, not just more agent branding.

At a glance. Here is everything else worth knowing from today’s AI news.

• OpenAI published an Endava Codex case study. OpenAI says Endava is using Codex to shorten requirements analysis, a claim businesses should test against their own delivery data.
• Apple’s next Siri redesign is still a watch item. TechCrunch reports that new renders point to a standalone Siri app and broader assistant overhaul.
• Hy3 is drawing attention on OpenRouter rankings. Max Woolf’s write up argues the model’s sudden ranking lead deserves scrutiny before anyone treats it as settled evidence.
• AI token futures are moving from idea to market design. TechCrunch reports that exchanges are looking at derivatives around AI token use, which could turn compute demand into a traded input.

The forward signal is clear: agents are becoming less like experimental chat windows and more like connected software workers. The next question is whether vendors can make their permissions, costs, and audit trails as visible as their demos.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.