Forward-deployed / Learning zone
Agentic workflowsa Generative AI module
A standalone module

Agentic workflows for the product leader

Orchestrating more than one agent when a single loop isn't enough, and what it takes to make a workflow durable enough — and valuable enough — to be worth owning end to end.

2 lessons+ recapknowledge graphdiagrams included

A single agent's loop only gets you so far. Some jobs are too big for one context window, too parallel for one worker, or too long-running to fit inside one continuous run — they need to pause for days waiting on a human, survive a restart, and pick up exactly where they left off. "Agentic workflow" is the name for what happens once a task needs more than one loop, or needs the loop to keep its place across real time. Getting this right is mostly a small number of orchestration and durability decisions, made deliberately instead of by whichever framework happened to be closest to hand.

A note on scope. This is the most heavily covered topic in this family so far. Multi-agent systems & protocols already develops orchestration topologies and the MCP/A2A protocol landscape in full depth. The workflow-versus-agent distinction itself is already the subject of AI agents's first lesson in this same family. Human-in-the-loop design is already developed as part of Agentic AI as a product's agent-UX section, and belongs more fully to this family's upcoming AI security & guardrails module. Durable, long-running execution is the entire subject of a dedicated twelve-phase track, Flowable, which builds a real process engine from scratch. Workflow capture as a business strategy is already a full section — and an existing glossary term — inside Agentic AI as a product. Re-deriving any of it here would only restate it a fourth or fifth time. This module compresses to the two lessons that are genuinely new once all of that is accounted for: how to decide on an orchestration shape once one agent isn't enough, and what it takes to make a workflow durable enough, and valuable enough, to be worth owning end to end.

The knowledge graph

Agentic workflows

Two questions asked in sequence

Once a single agent's loop isn't enough — what shape, and how does it survive real time?

Lesson 1Orchestrating more than one agent
Chains, routers, multi-agent topologies

The three recurring shapes work takes when one agent isn't the right unit.

Lesson 2Making it last
Surviving pauses & restarts

Wait states · persistence · job executor

→
Owning the whole workflow

The moat — one step vs. end-to-end

Durability is what makes the ownership strategy credible.

Read it as two questions asked in sequence. Orchestrating: once a single agent's loop isn't enough, what shape should the work take — a fixed pipeline, a router, several agents coordinating? Making it last: once that shape exists, does it survive being paused for three days waiting on a human, and is owning the whole thing — not just a step inside it — actually the better business bet?

The lessons

Each lesson pairs the product framing with a 🎯 For the product leader briefing — why it matters, the decision it changes, the question to ask your team, and the risk if ignored — plus a diagram. For the engineering depth behind every mechanic mentioned here, follow the spokes into Multi-agent systems & protocols, Flowable, and Agentic AI as a product.

📌 Close out the module: Recap & real-world examples.

The lessons

01

Orchestrating more than one agent

Read lesson →
02

Making a workflow durable, and worth owning

Read lesson →
📌

Recap & real-world examples

Read recap →