Insights
Insights
Notes on AI automation, SMB operations, and shipping workflows that survive contact with reality.
AI Act risk classification: how to decide if your AI workflow needs more control
The dangerous AI system isn't always the most advanced one.
Provider or deployer? The wrong answer derails your AI Act work.
The AI Act divides actors into roles: provider, deployer, importer, distributor. Which role you have determines which obligations apply.
AI Act and transparency: what must Swedish companies show from August?
August 2026 isn't just about lawyers and high-risk systems.
AI literacy is already an obligation, not a future item
Since February 2025, AI Act Article 4 applies: all providers and deployers of AI systems must ensure sufficient AI literacy among their staff.
From 21% to 95%: Why AI Analytics Is a Data Governance Problem, Not a Model Problem
Anthropic automated 95% of internal analytics queries with Claude. The secret wasn't a better model — it was the right data infrastructure around it. What that means for SMBs.
The first AI Act step isn't a policy. It's an inventory.
Most companies that start with the AI Act make the same mistake: they write a policy before they know what they use.
Agent-first does not mean human-last
The discussion about AI agents often lands in one of two camps: either everything should be automated, or AI is a threat to human work.
Content gets better when the agent gets performance data back
Most AI content workflows work in one direction: the agent writes, the team publishes, and then — silence.
A good AI case has an owner, not just an idea
'We should automate that with AI.'
The CMS API shouldn't give the agent publishing power from day one
We just connected an AI agent to our CMS. It can create blog posts, fill in SEO metadata, generate FAQ sections, and handle translations.
Approval loops are not brakes. They are design.
AI agents become dangerous when they get more rights than the process can handle.
The most dangerous AI agent is the one that almost works
The agent that gives the right answer nine times out of ten feels reliable. The tenth time, it sends an incorrect price to a customer. Or publishes a post with the wrong tone. Or updates the CRM wi...
We save the image brief before we generate the image
AI image generation has gone from experimental to a daily tool. But there's a catch: most people save the image — not the process.
Your CRM data doesn't need to be perfect for an AI agent to help sales
'We just need to clean up the CRM first, then we can automate.'
Stop losing leads — let an AI agent answer your phone
Missed calls cost small businesses thousands every month. An AI phone agent answers around the clock, builds trust, and collects everything you need to quote faster.
AI agents are not chatbots. They are workflows with responsibility.
The difference between a chatbot and an AI agent is not the model. It is the workflow, approval points, logging, and responsibility.
Claude Code’s Quality Drop Was a Product-Layer Failure, Not a Model Story
Anthropic’s Claude Code postmortem shows why strong models can still feel unreliable when the harness changes. The lesson for agent builders is simple: evaluate the runtime, not just the model.
Google's Restaurant Booking Rollout Shows What Agentic AI Actually Looks Like
Google's new AI Mode restaurant booking rollout matters because it turns AI from a better answer box into a narrow action layer. That is probably how agentic AI reaches the mainstream: through constrained tasks, live inventory, and partner-backed handoffs.
Claude Managed Agents Shifts the Bottleneck: From Prompts to Infrastructure
Anthropic’s Claude Managed Agents is a bet that the hard part of agentic AI isn’t model intelligence. It’s everything around it: long-running sessions, sandboxing, orchestration, and debugging. Here’s what changed, and what to do about it if you’re building production agents.
Your Competitor Just Hired an Employee That Never Sleeps
Swedish SMBs are sitting on a narrow window of competitive advantage. AI agents aren't a future thing. They're a right-now thing. Here's why waiting is the most expensive option.
Building AI Agents That Actually Work in Production
Most AI agent demos are impressive. Most AI agent deployments are disasters. Here's what separates the two: architecture patterns, evaluation strategies, and the pitfalls that kill production agents.
The Real ROI of AI Automation: What Actually Works and What Doesn't
AI vendors promise 10x productivity. The reality is more nuanced. Here's an honest breakdown of where AI automation delivers genuine ROI for small businesses, and where it's still burning money.