Introduction

In our August article, AI Value Beyond Efficiency: Build a Balanced AI Portfolio, we described three sources of AI value: Optimize, Redesign, and Create. With optimization, people use AI to draft, summarize, and analyze faster, while the underlying workflow stays largely the same.

This article focuses on Redesign: changing how the work itself gets done. Our July article, How to Build an AI Operating Model for the Mid-Market, covered the organizational foundations; here, we show how to redesign one workflow. That includes human-agent teams: people and AI agents handling different parts of a workflow, with people remaining accountable for the outcome.

At Escalate Group, we recommend starting with one workflow and a clear business outcome. Using a familiar example, the sales proposal process, we will illustrate five steps to divide the work, define approvals, build the necessary skills, and test whether the redesigned process delivers a better result.

What is covered in this article

• The difference between optimizing a task and redesigning a workflow.

• Five steps to redesign work for human-agent teams, using one proposal example.

• How to start small and decide what’s worth scaling.

• Frequently asked questions.

Optimize vs. Redesign: Where the Work Changes

Consider a sales proposal. With optimization, an account manager uses AI to prepare a draft, then follows the existing review process. With redesign, an agent could gather approved account information, check that the brief is complete, and prepare a draft for review. The account manager owns the client recommendation; finance reviews pricing or terms that fall outside agreed rules.

The opportunity is to change the sequence, handoffs, and decisions around the draft. Faster writing is useful. Removing an unnecessary review loop may matter more to the customer.

At Escalate Group, we recommend starting with this question: “If we redesigned this workflow today, knowing what AI can now do, how should the work be divided between people and AI?”

Five Steps to Redesign Work for Human-Agent Teams

For mid-market leaders, we recommend approaching workflow redesign as a practical sequence. The proposal process below is an illustrative example, not a client case study. Use it to see how the decisions connect, then apply the same questions to your own workflow.

Step 1: Choose One Workflow and Define the Outcome

Choose a recurring process with a clear owner, a measurable result, and enough friction to justify changing it. Other candidates include customer onboarding, supplier orders, and month-end reporting. Involve the people who do the work.

Before designing anything, agree on the outcome. For our proposal example, the aim might be to get an accurate, approved proposal to a qualified customer sooner, while giving the account manager more time for the client conversation.

Map the process from a qualified opportunity with an agreed brief to delivery of the approved proposal. Include information gathering, pricing, drafting, review, and client delivery. Record who does each part, which systems they use, and where work waits or comes back for correction.

Look closely at the delays. Is pricing hard to find? Is the customer brief incomplete? Are several people checking the same thing? Simplify those issues before introducing an agent. Agree on which record is authoritative when information conflicts.

Leave this step with a current workflow map, one accountable owner, and a baseline for the outcome you want to improve.

Step 2: Decide What People Own and What Agents Can Handle

Use the map to decide how to divide the work. Consider the quality of the available data, the clarity of the rules, and the consequence of an error. A task that looks repetitive may still need human judgment when the inputs are incomplete.

Agent-led: gather approved account information, check required inputs, and assemble a draft using current templates and pricing sources.

Shared: the agent prepares the proposal; the account manager checks accuracy, relevance, and the recommendation.

Human-led: clarify customer needs, decide what to offer, negotiate and approve commitments.

In this example, the agent returns missing scope information to the account manager before drafting. Finance handles exceptions to approved pricing or terms. The account manager approves the final proposal and sends it to the customer.

A well-designed human-agent workflow still has clear human accountability, even when agents perform much of the work. Name the person who owns the result and who can change the workflow when it needs attention.

Step 3: Set Permissions, Approvals and Escalation Points

Agents are becoming part of everyday business software. In a July 2026 update on Microsoft Copilot, Microsoft describes its Scout autopilot agent as one that “stays active in the background, with its own identity and permissions, to keep work moving.” When agents work this way, access design becomes a leadership decision, not just an IT setting.

Write down the boundaries before connecting the agent to live work. Make them specific enough that the team can tell when an action is allowed and when it needs approval.

Access: which customer records, documents, and pricing sources can it read?

Actions: can it draft, route work internally, or update a record?

Approvals: which actions need a person’s decision before proceeding?

Exceptions: when must it stop, who receives the issue, and what happens if that person is unavailable?

Traceability: how will the team review its sources, actions, and corrections?

For the proposal workflow, start with read access to approved sources and permission to prepare a draft. Require the account manager’s approval before client delivery. Send discounts outside the approved range or changes to standard terms to finance. The company defines those limits; the agent applies them.

If the current price list is unavailable or the brief conflicts with the account record, the agent should flag the issue and stop that part of the work. Keep a manual route available so an exception does not leave the customer waiting indefinitely.

Begin with a narrow scope. Expand authority when testing and live results support it, and the owner has agreed to the change. This is how controls become part of daily work.

Step 4: Build the Skills Around the Redesigned Work

When the workflow changes, people need to learn how to manage their part. Prompting is foundational: the quality of an agent’s work depends on the instructions and context people provide. On top of that foundation, teams need the ability to check sources, review outputs, and handle exceptions.

For the proposal workflow, account managers need to provide a clear brief, verify the information used, and recognize when a draft misses the customer’s needs. The process owner reviews recurring exceptions and decides what needs to change. A designated builder, supported by IT or a partner where needed, maintains the connections, templates, and rules.

Practice with past proposals before going live. Include an incomplete brief, outdated pricing, and an unusual customer request. Ask the team to explain what they would accept, correct, or escalate. Use those exercises to improve both the workflow and the training.

Step 5: Test the Whole Workflow Against the Outcome

If redesign is judged only by hours saved, it will look like optimization and be funded like optimization. Return to the business outcome agreed in Step 1 and compare the redesigned workflow with the baseline.

For the proposal example, track a small set of measures:

Response time: elapsed time from a complete, qualified brief to an approved proposal sent to the customer.

Quality: proposals returned for correction, including pricing or scope errors.

Coordination: review rounds and time waiting for approval.

Capacity: proposals completed per account manager, alongside time available for client conversations.

Also review exceptions and any actions outside the agreed rules. Track proposal conversion over a longer period, but interpret it carefully: qualification, pricing, and customer decisions also affect the win rate.

Test on past opportunities first, then run a limited live pilot with human approval. Set review dates based on proposal volume so the team has enough completed cases to judge the result. Count the time and cost of reviewing and correcting agent output.

If drafting gets faster but finance approvals remain the bottleneck, revisit the review rules. If rework increases, address the inputs or draft quality before expanding. Use the evidence to decide whether to continue, adjust, or stop.

Start Small, Then Scale What Works

Mid-market leaders can begin by bringing the workflow owner, the people doing the work, and the relevant technology support into one conversation. Keep the first test narrow enough to learn from real cases and meaningful enough to matter to the business.

Once the workflow produces a better outcome, capture what can be reused: approved data connections, review standards, escalation rules, and training examples. Use those lessons to choose the next workflow. This connects a focused experiment to a repeatable way of working.

Conclusion: 

Using AI inside today’s workflow is optimization. Redesign begins when the workflow itself changes.

The opportunity is not to deploy the most agents. It is to redesign the right work.

Start with one workflow that matters. Bring its owner and the people who run it together. Agree on the outcome, map the delays, divide the work, and set the approval rules. Then test whether the new process delivers a better result for the customer and the business.

If you would like support identifying and redesigning a high-value workflow, explore Escalate Group’s AI Studio. We help leaders connect AI experimentation to measurable business outcomes.

Frequently Asked Questions

What are human-agent teams?

Human-agent teams combine people and AI agents within a workflow. Agents handle defined activities, such as gathering information or preparing drafts. People set objectives, make consequential decisions, and remain accountable for the outcome.

What is the difference between optimizing and redesigning work with AI?

Optimization makes an existing task faster. Redesign changes the workflow: who performs each activity, how information moves, where approvals happen, and how exceptions are handled. For example, AI can draft a proposal faster; redesign also changes how the team prepares, reviews, and approves it.

Which workflow should we redesign first?

Choose a recurring workflow with a clear owner, accessible data, measurable performance, and a problem worth solving. Check where the delays and errors occur before selecting the technology. Start with a scope the team can test and manage.

Do employees still need prompting skills if agents handle more work?

Yes. Prompting is foundational because an agent’s work quality depends on clear instructions and context. Employees also need to verify sources, review outputs, and handle exceptions. Process owners and builders need the skills to maintain the workflow and assess its performance.

How do we keep control when AI agents act on their own?

Define what each agent can read, change, and execute. Set human approval points, escalation routes, and a record of actions. Test exceptions as well as routine cases. Expand authority only when the results and controls support that decision.