Topic
AI Agents and Agentic Workflows
An AI agent is a system where a model decides what to do next, calls tools or APIs, observes the result, and repeats until a goal is met. That autonomy is what makes agents useful and what makes them fail. Most agent incidents in production trace back to tool execution and orchestration, not to the model.
This is where we spend most of our engineering. The pieces that decide whether an agent ships are unglamorous: retries and fallbacks on every tool call, an evaluation harness that runs before go-live, confidence thresholds that route to a human, and audit trails on every decision. The writing below is what we have learned building and running them.
What we would tell you on a call
Most agent failures are tool failures
The model is rarely the thing that breaks. An agent calls an API that times out, returns a shape nobody expected, or succeeds twice because the retry was not idempotent. Design the tool layer as you would any distributed system and most incidents disappear.
Autonomy is a dial, not a switch
The useful question is not whether the agent is autonomous but what it is allowed to do without asking. Route low-confidence decisions to a person, keep the high-volume routine ones automatic, and move the threshold as the evaluation data earns it.
An evaluation harness before go-live, not after
A test set of real cases, run in CI, is what lets you change a prompt or a model on a Tuesday without finding out on Friday. Teams that skip it end up unable to change anything safely, which is a worse outcome than a slightly weaker agent.
Audit trails are a product requirement
In regulated work the agent has to be able to show what it did and why, at the time it did it. Retrofitting that is expensive; writing every decision to an immutable trail from the first sprint is not.
Everything we have written on this
AI Agents vs Simple Automation
The AI agents vs automation decision is not about which is more advanced. It is about which one fits the problem. Here is how we decide, and what it looks like in production.
AI Customer Support Automation
AI customer support automation is easy to demo and hard to trust. Here is what autonomous resolution actually requires: grounding in current policy, capped action authority, and a clean path to a human when the agent should not act.
Agentic AI vs AI Agents, Explained
Agentic AI vs AI agents, in plain language with simple flowcharts. An AI agent is one AI worker that can use tools. Agentic AI is a whole system that plans, coordinates, and corrects itself to reach a goal.
Agentic Workflow Automation in the Enterprise
A field guide to agentic workflow automation in the enterprise: where autonomous agents genuinely earn their keep, the four failure modes that quietly break them in production, and a readiness checklist before you fund a build.
Multi-Agent Systems That Survive Production
Multi-agent demos exploded in 2023. The ones that survived contact with production all had the same traits: supervisor patterns, typed shared state, and ruthless failure handling.
Human-in-the-Loop Workflows for AI Agents
Between 2022 and 2025, the most successful AI teams stopped talking about replacing humans and focused on designing great human-in-the-loop workflows instead.
Why Agentic AI Fails in Production
Most agentic AI pilots fail not because the model isn't smart enough, but because of three systemic issues: unreliable tool execution, missing evaluation harnesses, and no human escalation design.
Proof in production
Key terms
- AI agent
- A system where a language model chooses actions, calls tools or APIs, observes the results, and repeats until a goal is met or it stops. The distinguishing feature is that the model decides what to do next, rather than following a fixed script.
- Agentic AI
- A broader description of systems that pursue goals over multiple steps with some autonomy, usually coordinating several agents, tools, and checkpoints. Agentic AI describes the overall approach, an AI agent is a component within it.
- Orchestration
- The layer that controls how a multi-step AI process runs: what happens in what order, what state carries between steps, what to do on failure, and when to stop.Most production agent incidents trace back to orchestration and tool failure handling, not to the model.
- Human in the loop
- A design where a person approves, edits, or overrides an AI decision before it takes effect, usually on the actions that are expensive or irreversible.
The service behind this
AI Agents and Automation
Replace repetitive workflows with intelligent agents.
