We help leaders become their company's AI experts.
Our focus is on the *business* dimensions of Generative AI - what works, what doesn't work, how it is being used, and what you should know to transform your business.
Join our Executive Summary newsletter to get stay on top of the most important AI news for business in under 30 minutes a month, or join our free newsletter to get every post in your inbox.
#AI #GenAI #GenerativeAI #LLM #LLMs #OpenAI #ChatGPT #Anthropic
| Main site | https://ai-for.business |
| Free newsletter | https://ai-for.business/newsletter |
| Executive Summary newsletter | https://ai-for.business/executive-summary-monthly |
| AI case studies | https://ai-for.business/category/case-studies/ |
How do you actually run AI agents safely, visibly, and at scale inside the business?
OpenAI recently introduced its Frontier agent platform. The emphasis isn’t just on model capability — it’s on control and observability:
- A unified "single pane of glass" view of all agents
- Logging and traceability for debugging and audit
- Metrics for understanding agent use and performance
- Clear identities, permissions, and boundaries per agent
Salesforce just signaled a major shift: enabling agents to access Salesforce directly through APIs, MCP, and CLI.
This highlights two larger trends:
1️⃣ Just like with the web, SaaS is evolving toward dual interfaces:
- One for humans (UI)
- One for agents (APIs)
2️⃣ The business software architecture is also evolving
- SaaS tools shift from destinations → building blocks
- Work tasks driven by both humans and AI agents as one seamless system
https://www.salesforce.com/news/stories/salesforce-headless-360-announcement/
The web is becoming a place for both humans and agents, complete with tools and configurations that let your site be used by AI systems.
A simple new tool from Cloudflare checks if your site is ready for AI agents.
It will help you understand:
• Can your site be easily crawled?
• Is your content structured to be agent-friendly?
• Are interactions accessible beyond traditional UI?
• Do you clearly signal intent?
Check out the tool here:
A Wharton study recently identified four distinct AI adoption phases for companies:
1) Copy-and-paste - individuals have AI tools, but manage all data flows by hand.
2) Workflow pilots - data flow is somewhat automated, but stalls between people.
3) Read/Write AI - data crosses role boundaries; reliability improves with practice and effort.
4) AI-first - small teams build around business outcomes, and use AI from the start. Planning is key!
https://gail.wharton.upenn.edu/research-and-insights/beyond-copy-paste/
KPMG is shifting rote audit work to AI agents, with the expectation that AI agents could handle roughly 20–30% of a standard financial audit by 2029.
Along with cost and time savings, this means junior roles are shifting to:
- supervising agent outputs
- investigating exceptions
- applying judgment where more critical thinking is needed.
Anthropic just announced their next model, Mythos... by not releasing it. Instead, they announced that new model is unusually powerful at cybersecurity, capable of autonomously identifying a wide range of software vulnerabilities.
Standout findings:
- Critical vulnerabilities in every major operating system and web browser
- Vulnerabilities across common open source software
- Oldest vulnerability over *27 years old*
How do you track and control AI agents as they become part of your enterprise?
Authentication tools like Okta sit at the front door of your company’s software.
And, like employees, AI agents:
• Need specific, targeted access to systems and data
• May be run internally or sourced from outside providers
• Act on behalf of users, teams, and workflows
Is authentication the next step in AI agent visibility and control?
🔗: https://www.theverge.com/podcast/902264/oktas-ceo-is-betting-big-on-ai-agent-identity
Suno recently announced great new updates:
- Voices, enabling use of your own voice across different parts of a track
- Custom models, trained on your prior work
- My Taste, personalization based on preferences
While the barrier to using sound in marketing is dropping quickly, it makes a clear viewpoint more important. Differentiation will come from how intentionally it’s used, and during which parts of the customer experience.
Walmart recently found that conversion rates were 3x lower in ChatGPT versus from Walmart's own website.
Walmart's diagnosis: shopping isn't a single action - it's a journey.
If a customer gets a new TV, they also want cables, a mount, and more. ChatGPT enabled the initial purchase, but fragmented everything after.
So, Walmart is embedding their own chatbot into ChatGPT to bridge the shopping experience.
🔗: https://www.wired.com/story/ai-lab-walmart-openai-shaking-up-agentic-shopping-deal/
How does AI reshape the software "build vs buy" calculation?
The Wall Street Journal recently spoke with several companies about exactly this, and the dividing line was clear:
Risk and oversight.
- If AI outputs are reviewed, refined, and owned by humans → building becomes compelling
- If systems run autonomously or carry high downside risk → buying remains the smarter bet.
🎁🔗:
https://www.wsj.com/tech/ai/ai-software-business-stock-market-4b17b432?st=tbPcXe
