Blog
Notes on building agentic AI
Fifteen short reads, drawn from our internal Agentic AI Masterclass: vocabulary, architecture, governance, and two real production case studies.
Case Study: Automating a Job Search With an Agentic Workflow
A fork-and-own, Claude-native framework for job applications, built entirely differently from the TradingAgents case study — the contrast between the two is itself the lesson.
Case Study: How a Multi-Agent System Debates a Stock Trade
TradingAgents is a real, open-source multi-agent trading system — analyst team, structured bull-vs-bear debate, independent risk gatekeeping, and a five-tier decision — where nearly every idea in this series shows up in working code.
Data Sovereignty in the Age of Agents: What "Zero Data Retention" Really Means
"Zero data retention" gets used almost as often as a marketing phrase as a precise technical commitment. It helps to break "data sovereignty" into four distinct, concrete levers rather than one vague checkbox.
Retrieval-Augmented Generation as a Governed Tool Call — Not Magic
RAG is frequently presented as a black box. The more useful and accurate framing is much plainer: RAG is a governed tool call, and treating it that way answers a whole set of previously fuzzy questions.
Trust but Verify: Schema Enforcement, Structured Outputs & Agent Evals
An agent's output is only as useful as the next step's ability to parse it. Getting a model to reliably produce machine-parseable output is its own layer of engineering, with a genuine ladder of guarantee strength.
Keeping Humans in the Loop: Where Oversight Still Matters
A fully autonomous loop isn't always the goal. Almost any production agent that touches something irreversible benefits from a human checkpoint — and it's a more elegant mechanism than it sounds.
Build It Yourself or Rent It? Self-Hosted Runtimes vs. Managed Agent Platforms
You still own the agent's logic — the platform owns the operational burden around it. The honest question isn't which is better, but how much of your engineering time should go into infrastructure versus behavior.
What Is a "Harness"? The Hidden Layer Behind Every Agent
Every agent runs inside something that interprets its instructions and executes the resulting steps. A great deal of what looks like a difference in "how smart" a system is turns out to be a difference in harness design.
Deterministic vs. Non-Deterministic: When Should an Agent Decide for Itself?
For any given step, should the model decide which tool to call and when, or should that decision be pre-scripted? A well-built agentic system typically contains plenty of both.
Three Ways an Agent Reaches Beyond Itself: Tools, MCP, and A2A
A language model, on its own, only produces text. There are three distinct mechanisms for extending it beyond that — and choosing the wrong one for a given job creates real complexity later.
Memory That Persists: How Agents Remember Across Sessions
If a session ends, or gets summarized away to save space, is everything genuinely lost? For a well-designed agentic system, the answer is no — but only because memory is a deliberately separate mechanism from context.
What Is "Context" in an AI Agent — and Why Does It Run Out?
Context is everything the model can see on the current turn — nothing more. The discipline of agent design is, in large part, the discipline of managing what does and doesn't make it into that window.
Prompts, Skills, Commands, Agents: A Field Guide to the Jargon
Agentic AI has accumulated vocabulary faster than most teams have agreed on definitions. A short, plain-language field guide so a team can use the same words to mean the same things.
From Chat to Agent: Where Does Automation Actually Begin?
Every agentic system alive today started life as the same humble thing: a single chat turn. What changed isn't the fundamental unit of interaction — it's how much structure got wrapped around it.