AI agents
Building AI agents: when you need one, and the checks to design in from day one
A plain guide for teams planning to build an AI agent. The definitions and advice come from Anthropic's own engineering article as read on 10 November 2026.
Building an AI agent starts with checking whether you need one. Anthropic's engineering article separates workflows, where language models and tools follow predefined code paths, from agents, where the model directs its own process and tool use, and advises finding the simplest solution possible and adding complexity only when needed. For many applications it says a single model call with retrieval and examples is usually enough. So the first step in building an agent is to check whether you need one. If you do, plan the tools, the points where a person reviews and the conditions that stop it.
plain answer
What is the difference between a workflow and an AI agent?
Anthropic's article, published 19 December 2024, defines workflows as systems where language models and tools are orchestrated through predefined code paths. It describes agents as systems where the model dynamically directs its own processes and tool usage and keeps control over how the task is carried out.
In our experience the practical difference for a business is predictability. A workflow does the same steps each time, so it is easier to test and review. An agent chooses its steps, so it needs more careful limits.
Sources: Anthropic, Building effective agents (https://www.anthropic.com/engineering/building-effective-agents), read 10 November 2026.
how to
When should you build an AI agent instead of a simpler workflow?
Anthropic recommends finding the simplest solution possible and increasing complexity only when needed. It says agents suit cases where flexibility and model-driven decision-making are needed at scale, and that for many applications optimizing single model calls with retrieval and in-context examples is usually enough.
A test we use at Kriv: can you write the steps down in advance? If you can, build a workflow. If the steps depend on what the model finds along the way, an agent may fit, and the checks below matter more.
Sources: Anthropic, Building effective agents (https://www.anthropic.com/engineering/building-effective-agents), read 10 November 2026.
how to
What do you need to design when building an AI agent?
Three things stand out in Anthropic's article. First, tools: it says it is crucial to design toolsets and their documentation clearly and thoughtfully (we add: list what each tool can read and change). Second, human checkpoints: it describes agents that can pause for human feedback at checkpoints or when they meet blockers, so decide where a person must approve. Third, stopping conditions: it notes that it is common to include stopping conditions, such as a maximum number of iterations, to maintain control.
Add one more from your side: a written record of what the agent did, so a person can review it afterwards.
Sources: Anthropic, Building effective agents (https://www.anthropic.com/engineering/building-effective-agents), read 10 November 2026.
security
How do you keep a built agent under control?
We recommend giving the agent only the access its task needs, keeping a person accountable for it, and testing it on real examples before it touches live work. Anthropic also recommends extensive testing in sandboxed environments, along with the appropriate guardrails (Building effective agents, read 10 November 2026). Review a sample of its outputs on a schedule, and decide in advance what happens when it is wrong or stops.
For regulated work, ask how each of these is recorded. Governance is part of the build, not a step added at the end.
Sources: Anthropic, Building effective agents (https://www.anthropic.com/engineering/building-effective-agents), read 10 November 2026.
kriv role
Where Kriv helps
Kriv AI scopes the workflow, decides with you whether it needs an agent at all, builds it with review points and written limits, and trains your team to run it. Kriv AI has achieved the Select partner tier in the Claude Partner Network Services Track. Scope and price are set after a short discovery call.
Kriv did not write the Anthropic article and this page does not speak for Anthropic.
limits
What this page could not confirm
The Anthropic article carries a note that much of the tooling landscape has changed since its publication in December 2024, so check current documentation before you choose tools. We cited only this one article and did not test any agent framework.
Straight answers
Frequently asked questions about Building AI agents: when you need one, and the checks to design in from day one
What is an AI agent?
Anthropic describes agents as systems where the model dynamically directs its own processes and tool usage (read 10 November 2026).
Do I always need an agent?
No. Anthropic recommends finding the simplest solution possible, and says a single model call with retrieval and examples is usually enough for many applications (read 10 November 2026).
How do I stop an agent from running too long?
Anthropic says it is common to include stopping conditions, such as a maximum number of iterations (read 10 November 2026).
Can Kriv build an AI agent for us?
Yes. Kriv AI scopes, builds with review points and trains your team. Scope and price are set after a short discovery call.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call