The difference between an AI-native workflow and a workflow that simply incorporates AI.
Skip to content

Redesigning the work

TL;DR: AI made writing code cheap, so the constraint moved to everything around it: reviews, approvals, handoffs, and waiting.

  • The biggest gains come from redesigning the workflow, not speeding up one step.
  • Find the real bottleneck, which is usually coordination, not coding speed.
  • Rework the workflow before you add an agent.
  • Match each step to a person, an agent, or deterministic automation.


AI has changed the economics of writing software. Code, tests, and routine changes that once took days can now be generated in minutes. But the time between an idea and working software in production depends on far more than how fast code gets written. It depends on waiting: for a clarification, an environment, a review, another team, and for the handoffs where work sits idle in between. When creation gets faster and that surrounding work stays the same, the constraint simply moves to everything around the code.1

A useful way to find where to start is to ask what AI is breaking. If you use AI to write code and leave the rest of the path alone, the pressure just moves downstream. Engineering leaders describe the same pattern: writing code stops being the bottleneck, and the load lands on everything after it, from pipelines and testing to security review and the work of getting changes deployed and operated. Budgets and build capacity get consumed faster than anyone planned for. Review turns into the job no one wanted, as engineers spend their days checking output instead of designing systems. And where a process was already weak, acceleration amplifies it rather than fixing it.

Where the work strains or breaks points to where it needs to be redesigned.

Why redesign beats acceleration

The reflex when delivery feels slow is to speed up the part that looks slowest, which usually means pointing AI at writing code. Because writing code is a small part of the path from idea to production, speeding it up leaves most of that path untouched. Adding an agent to a workflow built for people accelerates a single step, and that step may not have been what was holding things back.1

The larger opportunity is to treat the workflow itself as the thing to redesign. Steps that existed to coordinate people, or to work around how long something took by hand, may no longer be needed. New steps, such as verifying an agent’s output, may need to be added. A workflow reshaped this way improves how the whole system performs rather than one step inside it.

This is harder than adopting the tool, and it is part of why results have been uneven across the industry. Technical systems can be rewired quickly, but the way teams are organized, how decisions get made, and where accountability sits all change slowly. Where AI has been added without reshaping the work around it, the outcomes teams cared about have often stayed flat.2

What redesigning the work involves

  • Find the real bottleneck: Locate where delivery actually slows, which is usually coordination between people and teams rather than the speed of writing code.
  • Map the team dependencies that limit agent speed: Agents speed up work inside a team, but changes still cross team boundaries to reach production. Mapping those boundaries shows you where agent-speed work will queue.
  • Rework the workflow around agents: Reshape the workflow before adding an agent, dropping steps that only coordinated people, adding the verification agents need, and making intent explicit so agents have something clear to act on.
  • Match the method to the task: Decide for each step whether an agent, deterministic automation, or a person is the right fit.

How this connects to the rest of the playbook

Redesigning the work comes first. Once a workflow is reshaped, the next questions are how people and agents share it, which Working with agents covers, and how to govern and scale it from one workflow to many, which Governing and scaling covers. The three topics build on each other, so the decisions you make here shape what the later topics have to work with.

// References

  1. Cochran, Timothy, and Bryan Finster, Tim Ottersburg, Scott Prugh, and Amy Willard. "When Coding Is Free, Where Is the Bottleneck? A Guide for Leaders Building Agent-Ready Delivery Systems." Enterprise Technology Leadership Journal 3, no. 2 (Fall 2026). https://itrevolution.com/product/when-coding-is-free/.
  2. Eslick, Ian, and Collin Fox, Amanda Graham, Suzette Johnson, Chuck Lafferty, Tom Kilcommons, Rosalind Radcliffe, Cat Swetel, Annie Vella, Sam Yeats, and Robin Yeman. "The Sociotechnical Architecture of AI-Enabled Enterprises: Principles, Patterns, and Practices for Human-Governed AI." Enterprise Technology Leadership Journal 3, no. 2 (Fall 2026). https://itrevolution.com/product/sociotechnical-architecture-of-ai-enabled-enterprises/.

/ Go deeper on this topic

Redesigning the work