Get tomorrow's brief in your inbox
Today: Anthropic launched Claude Science for drug discovery and announced plans to develop its own treatments for neglected diseases. Dropbox integrated Claude across its platform to keep AI work grounded in organized data. AI-powered browsers from Perplexity, OpenAI, and Opera are competing to become your digital assistant.
Anthropic is moving beyond selling AI to making drugs with it. The company launched Claude Science this week, an AI workbench for scientists that consolidates fragmented tools and datasets into one environment. But the bigger news is that Anthropic will use it to develop its own drugs, focusing on treatments for neglected diseases.
This puts Anthropic in the unusual position of competing with its own customers. The company is already selling Claude to biotech and pharma companies who use it for drug discovery, and now it's entering the same race. Anthropic joins AI-first drug companies like Insilico and Google DeepMind's Isomorphic Labs, plus Big Pharma companies building their own AI tools.
Here's what matters for small businesses: AI drug discovery is real, but it's still a long way from an approved treatment reaching patients. AI can suggest molecules, identify disease targets, and speed up research, but human oversight is required at every step. If you're in healthcare, watch this space for new tools that could help with research, data analysis, or clinical trial support. If you sell to pharma or biotech, expect your customers to be experimenting with AI across their entire pipeline. The Verge
Dropbox integrates Claude to prevent AI workflow fragmentation. Three new integrations let you connect Dropbox files directly to Claude: a Dropbox Connector for Claude, plus plugins for Claude Cowork and Claude Code. The practical use case is simple: ask Claude to analyze a design brief or summarize client feedback, and it pulls the file from Dropbox without you digging through folders. When Claude generates content, you can save it back to Dropbox with full version history. This is useful if your team already lives in Dropbox and you're tired of copying context between apps. The integration keeps your AI work grounded in organized, reliable data instead of scattered across chat logs. Available now for all 700 million Dropbox users. Small Business Trends
AI-powered browsers want to replace Chrome and Safari. Perplexity launched Comet, an AI browser that summarizes emails, browses pages, and sends calendar invites. It's available now for $200/month Perplexity Max subscribers (there's also a waitlist). The Browser Company introduced Dia, an invite-only AI browser that looks like Chrome but can perform tasks across any site you've visited. Opera launched Neon ($19.90/month), which can research, shop, and write code even while you're offline. OpenAI's Atlas browser lets you ask ChatGPT about search results and has an agent mode for completing tasks. All of these are betting that browsers will become assistants that act on your behalf instead of windows onto the web. If you're shopping for a Chrome alternative, wait until the dust settles. Most of these are early beta or invite-only, and pricing hasn't stabilized. TechCrunch
Zapier published guides for choosing CRMs, cold email tools, and autonomous AI CRMs. If you're shopping for business software, these are worth reading. The best CRMs for small business include Vtiger (product-based businesses), Flowlu (service-based), Zoho (scaling), HubSpot (inbound growth), and Clarify (AI-native). For cold email, Hunter is best for freelancers and beginners (extensive free plan), lemlist optimizes deliverability, and Instantly makes it easy to scale. Autonomous AI CRMs include Salesforce Agentforce (enterprise customization), HubSpot (marketing and sales alignment), and Pipedrive (visual pipeline management). The common thread: AI features only matter if the core workflow fits your business. Don't buy an AI CRM if the CRM itself is wrong for how you work. Zapier Blog
Learn how to conduct an AI agent security audit. If you're using AI agents that access customer data, financial records, or sensitive systems, you need to audit them before something goes wrong. The risks include leaked PII, prompt injection attacks, irreversible autonomous actions, and shadow AI (employees using unsanctioned tools). The audit process: (1) map what your workflow actually does (tools, triggers, inputs, decisions, actions), (2) audit user access and apply least privilege, (3) review what data the agent can touch, (4) test for prompt injection vulnerabilities, (5) add human-in-the-loop checkpoints for high-risk actions. With Zapier, you can use Canvas to visualize every connection and set asset permissions to control who can edit workflows. This is not optional if you're running AI agents in production. Zapier Blog
UiPath vs. Power Automate: which automation tool fits your business? Power Automate is best if your stack runs on Microsoft (Teams, Outlook, SharePoint, Excel). It's low-code, approachable, and easy to roll out across teams. UiPath is better for heavy, complex automation, especially with legacy systems that don't have APIs. It can handle desktop apps, no-API workflows, and structured enterprise automation programs, but the learning curve is steeper. The pricing reflects this: Power Automate is simpler and cheaper, UiPath scales with larger automation initiatives. If you want something that multiple teams can adopt without formal training, Zapier is easier to democratize while still giving IT visibility. Don't pick an automation tool based on features. Pick based on where your work actually lives. Zapier Blog
Fable token management tip: delegate coding tasks to lower-power models. If you're using Claude Code (Fable), you can preserve your Fable allowance by telling it to use its own judgment to delegate implementation work to Sonnet or Haiku subagents. The prompt: "For all coding tasks use your judgement to decide an appropriate lower power model and run that in a subagent." Design, auditing, data synthesis, and judgment-heavy work stays in the main Fable loop. Implementation gets delegated. This approach saves tokens without sacrificing quality because Fable still reviews the result before committing. If you're burning through your Fable allowance too fast, try this. Simon Willison
Macron and Modi are personally courting AI CEOs to secure infrastructure investment. French President Emmanuel Macron convinced SoftBank CEO Masayoshi Son to invest tens of billions into AI data centers in France by offering power capacity (nuclear) and government support. Macron also hosted AI bosses including Sam Altman, Dario Amodei, and Demis Hassabis at the G7 conference. India's Prime Minister Narendra Modi secured commitments from Amazon ($48B, $21B for AI and cloud), Microsoft (largest investment in Asia), and Google ($15B for India's largest AI hub outside the US). Both leaders are using personal relationships and long-term tax breaks to attract hyperscalers. What this means for your business: expect AI infrastructure to grow faster in countries offering power capacity and tax incentives. If you're building AI products, watch for regional data center availability to improve. CNBC
Current AI launched an Open Source AI Gap Map cataloging 421 AI products. The nonprofit, founded at the AI Action Summit in Paris (February 2025) with $400M in backing, released a map indexing the current state of open source AI. The map details 266 software tools, 85 models, 50 datasets, and 20 hardware projects across 14 categories and 3 layers (model components, product/UX, infrastructure). All data is released under an MIT license on GitHub as 1,184 YAML files plus notebooks and schemas. If you're evaluating open source AI tools, this is a useful reference. You can explore the underlying data using Datasette Lite. Simon Willison
Course creators report 50 percent revenue drop due to AI. Josh W. Comeau, who teaches web development, said his latest course is on track to sell one third as many copies as a typical launch. Sales across his existing courses are down significantly from last year. The reasons: (1) people are questioning whether developer jobs will exist, so they're reluctant to spend on learning, and (2) LLMs provide personalized tutoring for free. Other course creators are seeing the same trend. If you sell courses or training, expect AI to compress demand unless you offer something LLMs can't replicate (community, accountability, certification, hands-on projects). Simon Willison
Use natural language to query your data with AI-powered analytics tools. Retail analytics platforms like Zenlytic let you generate demand forecasts using plain English queries instead of SQL. This means non-technical team members can access insights without waiting for a data analyst. If you're drowning in spreadsheets and struggling to turn data into decisions, look for tools with natural language processing built in. The business impact: faster insights, better inventory management, fewer stockouts, and higher profitability. Start by identifying one business question you ask regularly (sales trends, top customers, inventory turnover) and find a tool that can answer it in plain English.
Audit your AI agent workflows before they break in production. Map every tool, trigger, input, decision, and action in your workflow. Review who has access and apply least privilege. Test for prompt injection by feeding the agent unexpected inputs. Add human-in-the-loop checkpoints for any action that touches customer data or financial systems. Use Zapier Canvas to visualize connections and set permissions. This saves you from discovering security holes after a data leak.
The pattern across today's stories is clear: AI is moving from experimental to operational, and that shift demands better governance. Anthropic is developing drugs, not just drug discovery tools. Governments are courting AI infrastructure like it's critical national infrastructure. Small businesses are adopting AI agents that can execute actions autonomously, which means security audits are no longer optional. If you're still treating AI as a side project, you're behind. The companies winning are the ones integrating AI into core workflows, auditing for risk, and building processes that can scale safely. Watch for more tools that consolidate fragmented AI workflows (like Dropbox's Claude integration) and more regulatory pressure on autonomous AI systems.