AI agents, answered.
Direct answers to the questions people actually ask about building and running AI agents. Each one starts with a short answer, explains the trade-offs, and lists the published sources it relies on.
AI agent orchestration
How the steps around a model are arranged — what runs next, what is fixed in code, and what the model is allowed to decide for itself.
What is AI agent orchestration?
AI agent orchestration is the logic that coordinates models, tools and agents: what runs next, what context each step gets, and how results are checked.
What is the difference between an agentic workflow and an AI agent?
In a workflow, code fixes the steps and a model performs some of them. In an agent, the model picks its own next step. How to tell which one a task needs.
What are the main AI agent orchestration patterns?
Prompt chaining, routing, parallelisation, orchestrator-workers, evaluator-optimizer and autonomous agents: what each pattern is for and when to use it.
When is an AI agent better than traditional automation?
An AI agent beats rule-based automation when a task needs judgement, has rules too tangled to maintain, or depends on unstructured text. Otherwise, it doesn't.
Multi-agent systems
When splitting work across several agents helps, what it costs, and the failure modes the published research keeps finding.
When should you use multiple AI agents instead of one?
Start with one agent. Split into several when the work is broad and parallel, or one agent has too many tools. What the published evidence says on both sides.
How do AI agents communicate and share state?
Inside one application, agents share a state object or hand off messages. Across systems, they use protocols such as A2A. What to share, and why it matters.
Why do multi-agent AI systems fail?
A study of 1,600+ traces found 14 failure modes in three groups: system design, inter-agent misalignment and task verification. Causes and mitigations.
How do you run several AI coding agents in parallel without conflicts?
Give each coding agent its own git worktree and branch, split work along file boundaries, agree interfaces first, and merge through normal review and tests.
Agent frameworks and protocols
The libraries agents are built with and the open protocols they use to reach tools and each other — described from their own documentation.
How do you choose an AI agent framework?
First decide whether you need a framework at all. Then compare control, state and durability, human-in-the-loop support, language, and model lock-in.
What is the Model Context Protocol (MCP)?
MCP is an open protocol that lets AI applications connect to tools and data through one standard interface. How it works, and what it does not do for security.
What is the difference between MCP and A2A?
MCP connects an AI agent to tools and data. A2A connects agents to other agents. Why the A2A project calls them complementary, and when you need each.
AI agents in production
What it takes to run an agent for real users: knowing whether it works, seeing what it did, limiting what it can do, and what it costs.
How do you evaluate an AI agent?
Build realistic tasks, run the agent several times on each, grade outcomes with code, models and people, and track capability and regressions separately.
How do you monitor AI agents in production?
Trace every agent run — model calls, tool calls, hand-offs — with inputs, outputs, tokens, latency and errors, then add outcome metrics and human review.
How do you stop an AI agent from taking actions it shouldn't?
Limit an agent's tools, permissions and autonomy, enforce authorisation outside the model, and require a person's approval for high-impact actions.
What drives the cost of running an AI agent?
Agent cost is mostly tokens: how many model calls a task takes, how much context each carries, and which model runs it. How to measure it and bring it down.
How do AI agents manage memory and context?
A model sees only its context window, so agent memory is engineered: saved state, compaction, notes kept outside the context, and clean-context sub-agents.
How should an AI agent recover when a step fails?
Make agent runs resumable: save state after each step, retry only what is safe to repeat, cap retries, and hand the task to a person when limits are reached.
Looking for decisions before a website or product is built? The articles cover those.
Tell us what you are building.
A short brief gets you a real reply from a founder within one business day — an honest read on fit, scope, and budget.