AI Agents
AI agents are software systems that can perceive their environment, make decisions, and take actions autonomously. They range from simple chatbots to fully autonomous systems. Understanding the spectrum of agent capabilities — and their limitations — is essential for leaders evaluating agentic AI for their organizations.
Work with Amir on AI AgentsBusiness frameworks for AI adoption
AI agent design patterns: a library for how people work with agents
I built a library of more than 30 AI agent design patterns in seven families, each traced to a product or research paper, plus eight recurring failure modes.
Running AI work wide, not just long
Most of the conversation about capable models is about jobs that run longer. There is a second direction nobody talks about: the same job run many times over, at once. It works, and it breaks in a place you would not expect.
AI for deep research: what the Stanford Virtual Lab still teaches us in 2026
In November 2024 a Stanford group spun up a Virtual Lab of AI agents using GPT-4o and ended up with experimentally validated nanobodies that bind a new SARS-CoV-2 variant. Eighteen months later the model looks weak and the design choices look stronger than ever. Four moves are worth pulling out: the AI picked its own team, five parallel meetings replaced individual judgement, the agents had hands, and compute time replaced calendar time.
Company as intelligence: from AI overlay to AI-native
Jack Dorsey's Block laid off 40% of its workforce and the stock went up 26%. His manifesto argues that AI breaks a 2,000-year-old hierarchy problem. Three structural layers, three surviving roles, and one diagnostic question your leadership team should be asking.
AI Disruption: Two Lenses for Seeing What's Coming
Disruption means questioning whether your industry's operating model will still make sense in three years. Two lenses — value chain compression and new actor emergence — help you see where it's heading.
Rethinking Engineering Organizations: The Block-Coinbase Contrast
Two companies, two radically different approaches to AI in engineering. Block cut 40% and called it AI transformation. Coinbase shipped 3-4x faster without losing anyone. Which path creates lasting value?
Intent Engineering: The Missing Layer in Enterprise AI
We've taught AI what to know. We haven't taught it what to want. That gap is why most companies still see no tangible value from AI - and the fix starts with something Peter Drucker told us decades ago.
Building AI Agents That Work: 10 Design Principles for Business Leaders
Ten practical principles for leaders building or evaluating AI agent systems - whether for customer experience, internal operations, or any business function.
Agent Teams: The Next Frontier of AI Business Value
I ran two sessions this week with AI champions at a customer I'm working with. The topic: what agents actually mean, and what the role of humans becomes when working with them.
The Five Levels of AI Autonomy
Everyone's racing to build "AI agents," but most companies are thinking about this wrong.
Leadership principles for the AI era
Team Tenets for humans and agents: writing down what a team believes
Team Tenets help a team make decisions faster. Here is what changes when AI agents join the team, with three example tenets for a hybrid team.
Ownership: build one AI frontier team before you scale
I think now is the time to build one team that experiments with AI. Three or four people, real freedom to fail, an outcome they own, and AI agents as part of how the team works. Start with one, and let it teach you how to scale.
Think bigger: what long-horizon AI models ask of leaders
The newest models stay on a single job for hours, sometimes days. The capability is obvious once you use them. Working out what to hand over turns out to be the harder problem, and it is a leadership problem more than a technical one.
AI as Aliens: what Project Hail Mary teaches us about working with other intelligences
In Project Hail Mary, the human and the alien start off without a shared language. They build one out of math, physics, and patience. The collaboration only starts to work when each understands what the other is built to do. That posture, sustained curiosity about a partner who genuinely does not see what you see, is the posture that might make AI work actually pay off in 2026.
Six thinking hats for hybrid teams
Edward de Bono's 1985 framework for separating thinking modes turns out to be the most practical guide for leading teams where some members are human and some are AI. Parallel thinking is what agentic AI has been quietly implementing all along.
The Jazz Model: Leading Hybrid Human-AI Teams
Leaders managing hybrid human-AI teams need a new mental model. Jazz ensembles — not orchestras — offer five principles for leading teams where humans and AI agents improvise together.
Team Tenets: From Leadership Principles to Practical Mechanisms
Leadership principles inspire direction, but tenets resolve the real tradeoffs your AI team faces every day. Here's how to write team tenets that accelerate decisions and align autonomous systems.
Bias for Action Revisited: When Experimentation Cost Approaches Zero
Five months ago, the question was whether to try AI. Now experimentation costs have collapsed — what happens when bias for action meets near-zero cost iteration?
Leadership Principles in the Age of AI Agents
The same principles that built one of the world's most innovative companies are now the playbook for leading AI-powered organizations. Here's how seven of Amazon's leadership principles apply when your team includes agents.
Strive to Be Earth's Best Employer in the Age of AI
It's been a tough few weeks for some of my former AWS colleagues.
Hire and Develop the Best: Leading Hybrid Human-AI Teams
This month I onboarded a new team member. I thought carefully about what context they'd need to succeed - our goals, our working style, the projects in flight, where to find key documents.
AI tool deep-dives and reviews
Instinct and the new personal agents: what they mean for your business
Personal agents like Instinct now book and pay on their owners' behalf. What IHG had to work out when they arrived, and how your business can get ready.
SciSpace: AI-powered literature research for the rest of us
The Stanford Virtual Lab sits at the ambitious end of AI in science. Most working researchers and most executives trying to understand their R&D function will meet AI through something much more modest first. SciSpace is a good example: 280M papers, chat with your library, ~550 prebuilt task agents, inline citations as you write. The leverage is in pointing it at your own corpus first.
Google Veo 3.1: the jagged edge of AI video
Released by Google DeepMind in January 2026, Veo 3.1 is a text-to-video and image-to-video model integrated across Gemini, Flow, YouTube Shorts, and Vertex AI. The creative iteration speed, not the rendering quality, is the actual product.
Writing skills: teaching AI how you think, write, and create
Skills are no longer just a coding trick. With 740,000+ skills across 30+ AI agents, they're the open standard for encoding expertise. Three open-source skill projects show what's possible for writers, thinkers, and creators.
Base44 Superagents: Always-On AI Agents Without Code
The shift from AI that talks to AI that does is the defining trend of 2026. Base44's Superagents bring always-on, persistent AI agents to a no-code environment — here's what I've learned so far.
Comet Browser: AI-Native Browsing from Perplexity
Perplexity's Comet browser turns browsing from searching into understanding. With a free agentic mode and cross-tab awareness, it's the most accessible AI browser — but privacy and security concerns are real.
Base44 Academy: Building Apps Without Code
If 2026 is the year of agent teams, what happens when you apply that concept to software development itself?
Google AI Studio and Gemini 2.5 Pro: Enterprise AI Powerhouse
When Mika writes about paradigm shifts - about recognizing when optimization stops being enough - he's describing something happening right now in one of the most consequential business tools ever built: Excel.
AI terminology that actually matters
What are proactive AI agents?
Proactive AI agents start the interaction themselves. They watch your context and act or reach out to you before you ask, within limits you set.
What is recursive self-improvement (RSI)?
Recursive self-improvement is the point at which AI systems can design and build their own successors with little human involvement. In the past few weeks it moved to the centre of the industry's agenda.
What is a self-driving lab?
A self-driving lab is a research operating model in which AI proposes the next experiment, robotic instruments run it, ML models read the result, and the loop runs again with light human supervision. The term has been in scientific use since 2018. What changed by 2026 is that the lab-in-a-box version has moved from research-paper claim to routine operating mode, and now interlocks with multi-agent systems in a way that was not possible five years ago.
What is a skills-based organization (SBO)?
A skills-based organization (SBO) is the structure of work, mobility, and decisions organized around discrete skills rather than jobs and ladders. In 2026 the concept matters twice over: skills are the foundation for AI agents as much as for humans. Skills work pays twice.
What Is an Agent Harness?
If the AI model is the brain, the harness is the body — the infrastructure layer that connects thinking to doing, and the make-or-break factor for agents in production.
What Is Computer Use?
Computer Use is the AI capability to see, interpret, and interact with computer screens like a human — clicking buttons, filling forms, and navigating applications without needing APIs or custom integrations.
What Are MCP Apps?
In Issue #12, I introduced term MCP - the open standard that acts as a "USB-C port for AI," letting AI models connect to external tools and data sources through a universal interface. Since then, MCP has grown rapidly.
What Is AI Orchestration?
This Week's Term: Orchestration - the coordination of multiple AI agents, tools, or capabilities to accomplish complex tasks that no single component could handle alone.
What Is MCP (Model Context Protocol)?
This Week's Term: Model Context Protocol (MCP) - an open standard that enables AI assistants to connect with external data sources and tools through a universal interface, allowing applications to provide context to ...
What Are AI Agents? The Five Levels
AI Agents are autonomous AI systems that can reason about problems, decide which tools to use, take actions across multiple steps, and iterate on their work without human intervention at each decision point, moving be...