🎯 “How do we pick the RIGHT AI agent use case?

This is the question I hear most from customers exploring agentic AI.

Here’s the mechanism I run through together with the customer:

The 4-Quadrant Evaluation

When a customer brings me 5-10 agent ideas, we structure each one across four dimensions:

📊 Business Value & Strategic Fit → What pain does it solve? For whom? How often? → Can we quantify the impact? (Revenue, cost, time, quality) → Which KPI moves if this works for 6 months?

Kiro Subagents: Scaling Development with Specialized AI Agents

When you’re building complex software, context management becomes your bottleneck. Your AI agent is juggling frontend components, backend APIs, database schemas, testing frameworks, and documentation—all competing for limited context window space. The result? Diluted focus and suboptimal outputs.

Kiro Subagents solve this architectural challenge by enabling parallel task execution through specialized, autonomous agents that maintain independent context windows.

🏗️ The Architecture: Parallel Contexts, Focused Execution

Subagents operate as independent processes with their own context management. This architectural pattern delivers several technical advantages:

🎯 From Chaos to Control: Building Predictable AI Agents That Get Smarter Over Time

⚖️ We need to balance Agency versus Control. We want AI systems to be super easy to use, read our minds, and just provide the answer we need. But we also need to make sure that nothing goes wrong. The more we control, the less agency we get. This is a balancing act.

Let’s focus on the control part. There are many different mechanisms to increase and guarantee control. Things like policies and guardrails come to mind. Those are obvious and powerful. I will cover them in a dedicated post.

😱 “Argh - My AI Agent deleted all my files!!!!”

When was the last time one of your co-workers deleted an important file from your desktop?

I would hope it has been a long time ago and quite possibly never.

🖥️ Even at the time we used shared computers, home directories have been separated.

There was a notion of shared folders or network folders or whatever nomenclature the system you were using. So in order to provide access to files to coworkers, you would need to take the decision first to upload it to a shared folder.

When a ‘Model’ Isn’t Just a Model: Redefining AI Systems for the Builder’s Era

🎬 Great keynote by Jensen Huang at CES 2026 [1]! Great content and also love the ease of his presentation style. Miguel: We are not the only ones presenting in front of a black screen once in a while ;)

🔓 I agree with Jensen, it’s super exciting to see more and 𝗺𝗼𝗿𝗲 𝗼𝗽𝗲𝗻-𝗶𝘀𝗵 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗺𝗼𝗱𝗲𝗹𝘀 𝗯𝗲𝗶𝗻𝗴 𝗽𝘂𝗯𝗹𝗶𝘀𝗵𝗲𝗱 by different providers. Sounds like NVIDIA is taking a big stake in this. Really key for me is that providers not “just” release open-weight models but also the data they trained on and the process used to train them. Jensen mentions the obvious responsible AI argument which is super important. This is the only way 3rd parties can verify the models and understand things like bias being introduced by the training data, copyright infringements, and alike. From my perspective, equally important: 𝗢𝗽𝗲𝗻 𝗶𝘀 𝗼𝗻𝗹𝘆 𝘁𝗿𝘂𝗹𝘆 𝗼𝗽𝗲𝗻 𝘁𝗼 𝗺𝗲 𝗶𝗳 𝗜 𝗰𝗮𝗻 𝗯𝘂𝗶𝗹𝗱 𝗶𝘁, 𝗺𝗼𝗱𝗶𝗳𝘆 𝗶𝘁 𝘁𝗼 𝗺𝗮𝗸𝗲 𝗺𝘆 𝗼𝘄𝗻 𝘃𝗮𝗿𝗶𝗮𝗻𝘁, 𝗮𝗻𝗱 𝗜’𝗺 𝗮𝗹𝗹𝗼𝘄𝗲𝗱 𝘁𝗼 𝗱𝗼 𝘀𝗼.