Skip to main content

Articles about AI & Automation

Honest analysis of AI applied to real operations. No hype, just judgment: what you control, what you can prove, and what to build that your competitors can't.

The Sandbox Blocked POST. The Agents Used GET.
AI & Automation
· 5 min read

The Sandbox Blocked POST. The Agents Used GET.

A 25-year-old Austrian wiki took 18,000 messages from OpenAI agents that were forbidden to write. The containment rule described the wrong mechanism.

AI agents OpenAI AI governance
Lemkin Swapped 10 Reps for 20 Agents and 1.2 Humans
AI & Automation
· 3 min read

Lemkin Swapped 10 Reps for 20 Agents and 1.2 Humans

Jason Lemkin replaced SaaStr's sales team with AI agents. The part nobody quotes is that each agent needed 30 days of training before it was useful.

Jason Lemkin SaaStr AI agents
Anthropic Deleted 80% of Claude Code's System Prompt
AI & Automation
· 4 min read

Anthropic Deleted 80% of Claude Code's System Prompt

Anthropic cut more than 80% of Claude Code's system prompt with no measurable loss on coding evals. The rules you added last year are now the ceiling.

Anthropic Claude Code context engineering
Thomson Reuters Trained Its Own Model for $450,000
AI & Automation
· 4 min read

Thomson Reuters Trained Its Own Model for $450,000

Thomson Reuters built a legal model that beats GPT 5.4 and Claude Sonnet 5 on its own content. The final training run cost $450,000. The corpus is the moat.

Thomson Reuters specialized models proprietary data
Pinecone Nexus Beat GPT-5.5 by One Point, at 77% Less
AI & Automation
· 4 min read

Pinecone Nexus Beat GPT-5.5 by One Point, at 77% Less

Coverage said Nexus outscored frontier models. The primary says 47.4% against 46.4%. What the knowledge layer actually bought was cost, on the cheaper model.

Pinecone Nexus Sierra
Uber: 50+ Approvals per Session, Zero Real Oversight
AI & Automation
· 7 min read

Uber: 50+ Approvals per Session, Zero Real Oversight

Uber open-sourced ADR after admitting its own tools could not see what its agents did. The finding: approving 50+ actions per session is not real oversight.

Uber ADR agent security
Anthropic gave Figma three days' notice before competing
AI & Automation
· 5 min read

Anthropic gave Figma three days' notice before competing

Anthropic's chief product officer resigned from Figma's board on April 14. Three days later, Anthropic launched Claude Design, its direct competitor.

Figma Anthropic Claude Design
AI Doesn't Make You Better. It Amplifies What You Are
AI & Automation
· 5 min read

AI Doesn't Make You Better. It Amplifies What You Are

An engineer with Claude closes in an afternoon what used to take a week. The same tool, in careless hands, wipes a production database instead.

AI agents AI governance Claude Code
The model is a commodity. Governance is the moat.
AI & Automation
· 5 min read

The model is a commodity. Governance is the moat.

Enterprise AI does not fail because the model cannot reason. It fails because nobody owns the control tower: who approves what, under which policy.

AI governance AI agents enterprise AI

Stay up to date with AI and digital transformation

Join our weekly newsletter with practical analysis and tips

We respect your privacy. Unsubscribe anytime.

Frequently Asked Questions about Our Blog