Claude 4 agentic capabilities
Anthropic's latest model generation is designed with agent tasks as its primary use case. In practice, it shows: the planning horizon is longer, tool use is more consistent, and the model recovers better from partial errors in multi-step jobs.
Previously, agents often went off track around steps 3–4. With Claude 4, we regularly run 15–20-step executions without human intervention. It's not exponentially better — but it is enough to make the categories of possible tasks dramatically broader.
Open models match GPT-4 for business tasks
qwen2.5:32b runs locally on an RTX 4090 or RTX 5070 Ti and performs at the level of GPT-4 Turbo for most business tasks — summarizing, classification, writing, coding, analysis.
The most important consequence: you can run a powerful agent without API costs and without a single character of your data leaving your network. For GDPR-sensitive operations — healthcare, law, accounting — this is a game changer.
You no longer have to choose between capability and control.
Open 32B models 2026
n8n + Claude API: multi-agent systems for SME
n8n Cloud in combination with the Claude API has become the standard stack for European SMEs who want to build agent workflows without becoming software developers. Popular use cases right now: weekly competitive analysis, lead follow-up, content calendars, and invoicing control.
What is driving adoption is the price. n8n Cloud starts from 0 SEK/month for low volume, and Claude API's input/output prices have dropped enough that a 100-step daily workflow costs less than 50 SEK per month.
The Swedish Authority for Privacy Protection (IMY): GDPR and AI agents
The Swedish Authority for Privacy Protection (IMY) published guidance in May 2026 concerning AI agents directly: when an agent logs personal data as part of its execution — e.g., saves email addresses, names, or behavioral data — a legal basis is required according to GDPR Article 6.
For most business scenarios, legitimate interest (article 6.1.f) applies. However, you must document the balancing of interests. An agent scraping public LinkedIn data needs a documented balancing of interests — it is not enough to simply state that the data is public.
Practical recommendation: run an internal GDPR audit of all agent workflows you have in operation. Focus on what is logged, how long it is stored, and whether there is a legal basis.
We are past the 'try-a-chatbot' stage. The companies that win in 2026 have identified a specific repetitive process and built an agent specifically for it.
Daniel Merthen
Want to read more? See AI agents: what they are and why they are changing everything — deep dive published during the week.