Lesson 4 of 7
Workflows you can trust
7 min read
Letting an agent decide everything is exciting — and a little scary. What if you want the power of AI steps, but a path you can count on every time?
Set the path, not just the goal
An agent that plans its own every move is flexible but hard to predict. A workflow is the opposite trade: a fixed path of AI steps you lay out in advance. It's more reliable and repeatable — you know exactly what will happen. Anthropic's guide "Building Effective Agents" names five patterns that cover most jobs.
A workflow trades some flexibility for a lot of predictability: fixed steps, the same shape every time. Pick the pattern that matches your job.
Five shapes that cover most work
Chaining runs steps in a line, each feeding the next. Routing sorts the request and sends it to the right handler. Parallel splits the work into branches that run at once, then merges them. Orchestrate-workers has a lead break a big job into pieces for workers, then combine the results. Evaluate-optimize loops a maker and a checker until the work passes.
You don't need a genius do-everything agent for most tasks — you need the right simple shape. Start with the simplest pattern that fits.
Reach for a full, free-roaming agent only when the task really is open-ended. If a fixed workflow can do the job, it'll be cheaper, faster, and easier to trust.
The shape of it
- —A workflow is a fixed path of AI steps — predictable and repeatable.
- —Five patterns: chaining, routing, parallel, orchestrate-workers, evaluate-optimize.
- —Match the pattern to the job; prefer the simplest one that works.
Support tickets arrive, and each should go to billing, tech, or returns. Which pattern fits best?
Continue in the app
Take the whole Agents & Tools course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.