Get tomorrow's brief in your inbox
Today: Anthropic released Claude Opus 4.8 with improved honesty, now flagging uncertainties four times more often than its predecessor. Asana acquired StackAI for $75 million to power AI-native workplace teams. AWS rebuilt its cloud infrastructure specifically for AI agents, launching OpenSearch Serverless with zero-scale compute.
Anthropic released Claude Opus 4.8 today, and for once, an AI lab is being honest about what they shipped: a "modest but tangible improvement" rather than hyping it as revolutionary. The real story here is what they improved.
Opus 4.8 is four times less likely to let flawed code pass without comment, and it actively flags uncertainties about its work instead of confidently claiming progress when evidence is thin. In evaluations, it had the lowest hallucination rate of any tested model by choosing to abstain on questions where it was uncertain rather than guessing. For businesses relying on AI for code generation or analysis, this honesty could save hours of debugging and prevent costly mistakes.
The model also introduces mid-conversation system messages, letting you steer the AI's behavior without restarting the conversation. If you're running agent loops or long coding sessions, this preserves cache hits and cuts input costs. Pricing stays at $5 per million input tokens and $25 per million output tokens, with a faster 1,024-token prompt cache minimum (down from 4,096).
If you use Claude for business work, especially coding or analysis where accuracy matters more than speed, Opus 4.8 is worth switching to today. The improvement is incremental, but it's in exactly the right direction: telling you when it doesn't know rather than confidently guessing wrong.
Anthropic release announcement Simon Willison's detailed analysis
Asana bought the no-code agent builder StackAI for $75 million as part of its push to become an AI-native workplace platform. StackAI lets you design agents that operate within existing business systems like Salesforce, Slack, and Google Workspace, pulling in data and automating workflows without coding.
If you're already using Asana for project management, this acquisition means you'll soon be able to build custom agents directly inside the platform. Think of it as Zapier meets AI: agents that can monitor project status, draft updates, or automatically reassign tasks based on workload. Asana's bet is that its deep integration into existing workflows gives it an advantage over standalone AI tools from OpenAI or Anthropic.
For small businesses, the promise is that you could offload repetitive project coordination to an agent rather than hiring an admin. Whether Asana can deliver on that promise at a reasonable price point remains to be seen, but the acquisition signals that workplace automation tools are moving from "AI features" to "AI-first design."
Gumloop is an AI-native automation platform focused on building agentic workflows. It connects apps, logic, and AI steps in a visual no-code builder, and it supports Model Context Protocol (MCP) for tool chaining. The catch? Its integration library is a fraction of Zapier's size, and its credit-based pricing can get expensive fast.
Every workflow costs at least 1 base credit, then each node adds more. A standard AI call costs 2 credits, an advanced AI call runs 20 credits, and enrichment nodes cost 60 credits each. Agent interactions cost even more. If you're experimenting with AI workflows and don't need to connect to your entire tech stack, Gumloop might be worth testing. If you need reliability and speed to value, stick with Zapier's AI orchestration across 9,000+ apps.
The lesson here: AI-first platforms are exciting, but if they can't integrate with the tools you already use, you'll spend more time building workarounds than automating work.
Oculus founders' new AI startup Sesame launched its iOS app today with four conversational AI agents (Maya, Miles, Simone, and Charlie) that talk like humans, even mid-sentence pivoting when they find new information. The app runs parallel searches while speaking, weaving results into responses in real time rather than making you wait for an answer.
It's free for now, with a short waitlist at sign-up. The app includes incognito mode for private conversations, texting mode when you can't speak aloud, and search cards with image results. The real vision here is intelligent eyewear launching in 2027, with agents that can take action on your behalf rather than just answering questions.
For business use, this is still early. But if you're tired of ChatGPT's stiff conversational style or want an AI assistant that feels more natural to talk through problems with, it's worth joining the waitlist. Android preview is coming later.
Adobe launched the Firefly AI Assistant in beta, a conversational agent that operates Adobe's design apps (Photoshop, Illustrator, Lightroom) for you. Instead of just generating images, it explains step-by-step what it's doing using established editing terminology, and it's surprisingly honest about its limitations.
When you ask it to edit a photo, it describes the scene in detail, mentions which tools it's using, and tells you when something isn't working. The results aren't perfect (colors can be too vivid, blending isn't always clean), but they're convincing at a glance. Think of it as a mediocre design intern: useful for busywork like background changes or color adjustments, not ready for professional-grade work yet.
For small businesses that can't afford a designer but need basic image edits for social media or ads, this could save hours. Just don't expect portfolio-quality results.
Endava, a global software contracting firm, now calls itself an "agentic organization" after adopting OpenAI's Codex. The key insight: they're using Codex not just to write code, but to codify senior engineering expertise so junior developers can produce senior-level work.
Senior architects record their design decisions and best practices, then Codex acts as a guide when junior teams are coding. They can ask questions about patterns they don't understand, and Codex teaches them better practices in real time. One senior's perspective, encoded into Codex, can guide multiple teams in parallel. Requirements analysis that used to take weeks now takes two one-hour meetings.
For small businesses with a few experienced people and a team of less-seasoned staff, this is a template worth studying. The ROI isn't just in writing code faster. It's in scaling expertise across your team without hiring more senior people or spending months on mentorship. If you're using Codex or Claude Code, ask yourself: are you just writing code faster, or are you capturing institutional knowledge that makes your whole team better?
A new study shows female solopreneurs are adopting AI tools faster than men and saving an average of 6 hours per week by automating tasks like drafting sales copy, mastering SEO, and managing social media. Solo founders are treating AI not as a novelty but as a vital tool to balance multiple roles without burning out.
If you're running a business solo or with a tiny team, the pattern here is clear: AI adoption isn't about staying cutting-edge. It's about not drowning. The founders seeing results are using AI for high-friction tasks that used to require hiring specialists (copywriting, SEO, content creation) and treating the time saved as a strategic asset.
Practical takeaway: identify one task this week that takes you 2+ hours and feels like busywork. Find an AI tool that can do 80% of it, even if the output isn't perfect. You'll get better at prompting, and you'll buy back time for work that actually grows your business.
Anthropic announced a $65 billion Series H financing at a $965 billion valuation, making it the most valuable AI company in Silicon Valley and surpassing OpenAI's $852 billion valuation. Revenue has exploded to a $47 billion run rate (up from $30 billion earlier this year) driven by Claude Code adoption.
For small businesses, this matters because Anthropic isn't going anywhere. If you've been hesitant to build workflows around Claude because you're worried about the company's longevity, that concern is off the table. The funding also suggests continued investment in making Claude more capable, which means the tools you use today will likely improve.
China's Shanghai Futures Exchange is designing a derivatives market for AI tokens, and major exchanges like CME Group and ICE are building futures contracts for GPU rentals. This infrastructure buildout reflects enterprise demand for hedging against the cost of AI compute.
For small businesses, this is a signal: AI pricing volatility is real, and financial markets are taking it seriously. If you're running significant AI workloads (like processing thousands of images daily or running agent-based customer support), start tracking your token costs now. When futures markets launch, you'll have data to make smarter decisions about fixed-price contracts versus pay-as-you-go pricing.
Amazon launched its next generation of OpenSearch Serverless, a search and vector database redesigned specifically for agentic workloads. The key technical change: it decouples compute from storage, allowing compute to scale up in seconds for agent traffic bursts and scale down to zero when idle. You pay $0 when agents aren't running.
This is important because AI agents behave differently than humans. They can unleash hundreds of parallel queries in seconds, then disappear. Previous infrastructure required you to keep at least one instance running (and paying for it) even when idle. Now you only pay for what you use.
For businesses building agents (customer support bots, research assistants, workflow automation), this pricing model removes a major cost barrier. You're no longer paying for idle infrastructure while testing or during off-hours. It's a shift from "always paying for a parking space" to "paying for metered parking."
CNN filed a lawsuit against Perplexity, claiming the AI search startup generates verbatim copies of its articles and provides users with content locked behind CNN's paywall. CNN joins The New York Times, Encyclopedia Britannica, Merriam-Webster, and The Wall Street Journal in suing Perplexity for copyright infringement.
For small businesses using Perplexity or similar AI search tools, this is a reminder: the legal landscape around AI-generated content is unsettled. If you're relying on AI tools to summarize or repurpose content for your business, make sure you're adding your own analysis and perspective rather than just republishing AI-generated summaries. The risk isn't just legal exposure; it's that your audience will notice you're regurgitating content rather than providing original value.
Bloomberg published renders of iOS 27 showing Apple's Siri redesign, including a new standalone app with a ChatGPT-style chatbot layout and a pill-shaped chat bubble that pops out of the Dynamic Island. Users will be able to start conversations with Siri anywhere in iOS 27, similar to Android's Gemini integration.
The new Siri app will include chat history, attachment buttons, and voice mode. The Camera app will reportedly add a "Siri" mode, and the Photos app will include new AI editing features. Apple will unveil the full overhaul at WWDC on June 8.
For Apple-using small businesses, this signals that Siri is finally becoming a real productivity tool rather than a voice command gimmick. If the execution matches the vision, you'll be able to use Siri as an actual assistant for research, drafting, and task management rather than just setting timers.
ReAct Agents 101: If you're building AI agents, adopt the ReAct pattern (Reasoning + Acting). Instead of pre-scripting every scenario, ReAct agents think out loud, call external tools (APIs, databases, search engines), evaluate results, and decide next steps. This reduces hallucinations because the agent grounds its outputs in real data rather than assumptions. Most importantly, set an iteration limit to prevent endless loops and runaway cloud bills. LangChain and LangGraph both support ReAct patterns out of the box.
Three patterns stood out today: AI labs are prioritizing honesty over hype (Anthropic's "modest improvement" messaging is refreshing), cloud infrastructure is being redesigned for agent-first workloads (AWS's zero-scale compute is the first of many), and workplace tools are shifting from "AI features" to "AI-native platforms" (Asana's $75 million bet on StackAI signals a fundamental rethink, not a bolt-on feature).
For small businesses, the takeaway is clear: AI is moving from novelty to infrastructure. If you're still treating it as an experiment, it's time to identify one high-friction workflow and automate it this month. The tools are ready, and your competitors are already moving.