# AI workflow An AI workflow is a preplanned sequence that uses AI for one or more steps while keeping the order under the firm's control. Source: https://rodamora.com/glossary/ai-workflow Updated: 2026-08-09 --- An AI workflow is a fixed sequence with one or more AI tasks inside it. A person decides the order before the run, not the model while work is moving. Gartner said in June 2025 that many use cases called agentic do not need agentic implementations, and it expects more than 40% of agentic AI projects to be canceled by the end of 2027. Across the 150+ franchise units in the delivery data I track, about 15% run trigger-fired workflows. This is the middle step many firms skip. A workflow gives you a written process, a named review point, and a result you can trace. It can still use an agent for one flexible decision. The path around that decision stays fixed. ## What an AI workflow looks like in delivery Take a proposal process. A call ends. The transcript arrives. One step pulls out what the client asked for and what concerns they raised. A second step drafts a proposal against the firm's pricing rules and past work. A third sends it to a named person for approval. A fourth files the approved version and updates the CRM. Four steps, always in that order. Two are AI tasks and two are system plumbing. A person sits at one gate. Nothing changes based on what the model found in the transcript. That fixed order is what makes this a workflow. The start matters too. A workflow begins on a trigger. That trigger might be a file appearing, a call ending, a schedule firing, or a status changing in the project tool. If a person still has to open a window and ask for the process, you have a chatbot with more steps. A production workflow starts on its own through [a named software role that starts work on its own](https://rodamora.com/blog/what-is-an-ai-employee). The industry definition matches this operator test. Anthropic's engineering guidance calls workflows systems where LLMs and tools are orchestrated through predefined code paths. It calls agents systems where LLMs dynamically direct their own processes and tool use. The fixed path is the boundary. ## Workflow or agent: how do you choose? Two questions settle the choice for most delivery work. **Are the steps the same every time?** If you can draw the sequence on a whiteboard and it holds for the next hundred runs, build a workflow. If the right next step depends on what the last step found, and you would need twenty branches to cover normal cases, an agent may fit better. Anthropic's guidance makes the tradeoff clear. Workflows give predictability on well-defined tasks. Agents give flexibility where model-led decisions are needed. **Can you check the output, and can you undo it?** Those properties of the task matter more than how clever the model sounds. Run that test before either build. It is the core of [delegate the inputs, own the outputs](https://rodamora.com/blog/delegate-the-inputs-own-the-outputs). Work that is hard to check and hard to reverse belongs in a workflow with [a person at a named step](/glossary/human-in-the-loop), whatever a vendor calls the product. A workflow that fails is often easier to investigate because the path is known. You can find the faulty step and fix it. An agent can fail in different ways on different runs. Finding the fault then becomes part of the operating cost. This helps explain why Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing cost, unclear value, and weak risk controls. The workflow is often the design that would have worked. ## Where firms get workflows wrong **They automate a process nobody wrote down.** A workflow is a written process with AI in some steps. If the process exists only in the head of your most experienced person, you are discovering it at production speed and paying for the discovery in client work. Write the steps first. A recording of someone doing the job once can give you a useful first draft. **They turn skilled people into approvers.** A bad design puts the model at each production step and leaves the person as a rubber stamp at the end. The work may move faster while the job gets worse. The people doing it notice. I wrote about this failure in [the output-validator workflow](https://rodamora.com/blog/ai-anxiety-and-the-output-validator-workflow). Put the human gate where judgment changes the outcome, often at scoping and client context, not only at the end where the choice is yes or no. **They count workflows instead of results.** A firm can build twelve workflows and change nothing about its economics. An April 2026 US Census Bureau working paper found 66% of AI-using firms rely on AI only to augment existing tasks. A February 2026 NBER survey of nearly 6,000 senior executives found nine in ten reported no impact on employment or productivity over the prior three years. Building workflows is not the same as delivering differently. **They skip the workflow rung.** Agents look better in a demo, so teams fund them first. Across the 150+ franchise units I track, roughly 80% still run chatbots, about 15% run trigger-based workflows, and about 5% run autonomous agents defined by conditions. Most firms trying to jump from the first group to the third have not built the written process the second group requires. ## What does it cost to keep one running? The expensive part is maintenance, not the first version. Every workflow depends on the systems it touches. Someone renames a CRM field, the proposal step starts filing to the wrong place, and nobody notices until a client does. Give each workflow a named owner and a check that fails loudly. If nobody is accountable for its correctness this quarter, it will be wrong by the next one. Monitor the final business record, not only whether each software step returned a green status. The other cost is time for the people who learn the new way while still delivering the old way. In the delivery data I've reviewed, training that sticks runs 8 to 10 hours per person spread over about 3 weeks. It mixes video, in-person sessions, live Q&A, and technical support. Compressing it into one afternoon can launch a workflow, but it does not create a durable habit. The workflow gets [quietly worked around](https://rodamora.com/blog/ai-projects-dont-fail-at-your-size-they-go-underground). That is why one workflow at a time works better. Finish it, assign its owner, watch the result, and repair the weak step. Six announcements at once create fatigue and half-maintained systems. ## Where this sits on the Delivery Model Ladder A rigid workflow that fires on a trigger is Stage 2, Augmented, on the [Delivery Model Ladder](/delivery-model-ladder). At this stage, the firm decides what delivery staff no longer do by hand, what the review checks, and who remains in the chain. Using a chatbot to draft the same proposal faster is Stage 0 or Stage 1. The task moved faster while its shape stayed the same. The workflow is the jump to Stage 2 because the firm rebuilt a part of delivery around a defined path. Stage 3, [AI-native](/glossary/ai-native) delivery, is where agents take defined slices of work and revenue stops tracking headcount in the old way. Firms reach that stage by compounding sound Stage 2 decisions. They do not get there by skipping the written process. ## Quick answers **Can a workflow contain an agent?** Yes. Many production systems use a fixed path and place an agent inside one narrow step with a defined output. That gives the work flexibility where it needs it and predictability everywhere else. **How is this different from ordinary automation?** Ordinary automation moves structured data between systems with rules someone wrote. An AI workflow has at least one step where a model reads unstructured material, such as a call transcript, email thread, or scanned document, and produces something no rule could produce. **What should a service firm build first?** Start with the process your most experienced person runs often and describes the same way twice. Choose work that is high volume, repeatable, and checkable by someone else. Do not start with the process that would impress a client or move money.