A Chief of Staff’s Guide to Maximizing ROI from AI Agents

How to leverage AI for super-productivity, not cheap replacements

Hi everyone! Coming to you live from The Future We Wrk For — thank you to Stephen + wrkspace for having me! (And if you don’t know Stephen yet, you should, and tell him I sent you. He’s the best kind of people.)

---

I’m super excited about today’s guest post — I always get questions on how to utilize AI, and there’s so much nuance on how to do it well. Luckily, we have Sawyer to walk us through it.

About Sawyer

Sawyer Middeleer is Chief of Staff at Aomni and an AI transformation expert. He helps organizations of all sizes from startups to Fortune 100 enterprises leverage AI more effectively in their operations.

Sawyer is offering 1 hour of complimentary AI advisory to Your Chief of Staff readers. If you’re a strategic operator looking for support with your own AI transformation, reach out to Sawyer to book time.

---

A Chief of Staff’s Guide to Maximizing ROI from AI Agents

Last year the AI agent hype machine roared to life, but actual results left a lot to be desired. There were probably more influencers talking about agents on LinkedIn than there were companies actually using agents in production. I even went mini-viral for dunking on Lattice’s CEO for suggesting that AI agents have a place in your org chart.

But as I write this in June 2025, the vibes have shifted.

AI agents are here to stay

AI agents are everywhere, and making a case for themselves in your organization that no operator should ignore. A step beyond prompt-and-respond chatbots, agents are LLM-powered entities that are capable of planning and interacting with their environments to achieve goals. These capabilities enable them to own complex workflows across a wide variety of functions.

This can sound a bit abstract, so here’s a concrete example: Claude Code by Anthropic is an agentic coding tool that lives in a company’s software engineering infrastructure.

It can edit files, write new code, and even build entire features end-to-end. Claude Code is also capable of testing programs it writes and collaboratively contributing to a shared codebase through Git actions. In other words, it’s an AI agent that works just like a real engineer.

AI agents are driving the next evolution of work

The promise of fully autonomous “AI workers” is compelling enough that we also see early agents finding traction in operations, legal, sales, customer support, and other functions. At a high level, the ROI case seems obvious: 24/7 output plus efficiency from automation equals a sure win.

For as much value as AI agents can bring to organizations, the pain of missing the mark on implementation can be equally severe. If not designed and deployed thoughtfully, large-scale AI agent rollouts can quickly land you in hot water with key stakeholders.

Amid all the hype, it’s important to remember that AI agents are just another way of accomplishing work and delivering value to stakeholders. Getting value out of AI agents requires the right balance of cross-functional partnerships, systems thinking and strategic creativity.

In other words, AI agents are 100% in your wheelhouse as a Chief of Staff.

It can be a lot to wrap your arms around, especially if technology strategy is only one of the 73 other things you have on your plate. So here are a few tips to making sure you maximize the ROI of AI agents in your organization.

How to maximize the ROI of AI agents in your organization

1. Start with a foundation of good data

I can’t tell you how many times I meet with a leader who tells me they want to re-build their go-to-market around AI, only to find out their CRM is in dismal shape and none of their teams share data with each other.

No agent put into this environment will perform anywhere near the peak of its powers.

Context - the information a generative AI system has access to - is critical for agents to do their work effectively. AI systems need good data in order to take the right actions, create good content, or whatever else they’re designed to do. The most impactful thing you can do is ensure that your CRM, ERP, HRIS or whatever other sources of truth are in your IT alphabet soup are clean and up to date.

The other kind of data you need for agents are process data. When agents are built, they’re given lists of programs they have access to, actions they can take, and data sources they can reference.

Without any resources describing how work should be done and what “good” looks like, agents have to guess, and this leads to mediocre results.

Making sure functional leads have SOPs written for workflows they plan to automate will make onboarding agents that much smoother.

2. Focus on superpowers, not just cost savings.

Here’s the elephant in the room - many leaders regrettably see agents as a quick path to ROI by reducing headcount. I don’t agree that this is the best approach.

Instead, the truly compelling ROI case comes from enabling existing teams to be far more productive than they could have been before.

Here’s what this could look like in practice:

1. A support team offloading the 50% of easiest tickets to AI agents, while giving more attention to the highest value ones.

2. A project management team that offloads follow ups and admin work to agents, so they can spend more time optimizing project delivery and cultivating stakeholder relationships.

The theme you’re probably seeing is AI as an enabler of super-productivity, rather than a cheap replacement for business as usual.

This can be achieved across an entire organization (I’ve done it, multiple times) but requires the partnerships, systems thinking and creativity that only a CoS brings.

3. Automate workflows, not roles or tasks

Jumping from the strategic to the tactical now — I’ve found that a lot of folks get stuck when trying to decide what and how to automate. Do you pick one teeny-tiny problem to start with as a pilot? Do you buy one of those agents that claims it can do an entire job function?

The best way to think about agentic automation is to start with a high level goal, like “Talent Acquisition’s job is to find, attract and hire the best talent for our company.”

Then break that objective down into workflows, such as:

1. Identify near-term talent needs

2. Create job post messaging

3. Align the posts with the needs of specific channels

4. Get approvals from the TA manager

5. Go live with the postings

6. Update relevant internal IT systems

That’s a pretty good high level spec for a single talent sourcing agent.

There are a lot of companies out there trying to sell “agents” that just do one tiny piece of this workflow, like writing job postings. That might have cut it in 2024 but the best agents nowadays are capable of completing much more complex projects.

Notice also how a human remains in the loop to oversee this process.

Also be careful at the other extreme if you try to replace a role entirely. You’ll probably find that there’s a lot that an experienced human does really well that you take for granted, and will miss dearly when something goes wrong.

4. Remember: AI isn’t like other software

A roadblock that many teams run into when deploying AI agents for the first time is how frustrating it can be when they don’t do what you want them to.

This is especially true when you’re working with teams like Ops and Finance who place a premium on accuracy and security. It’s the tradeoff you make with generative AI -- flexibility and generalized problem solving in exchange for occasional weirdness.

The more robust interventions for AI agent performance are actually not related to tech or data at all — they’re operational and cultural.

First, any AI agent in your business-critical systems needs some foundational guardrails to make sure the occasional unpredictability stays within a safe range. You’ll often hear these termed:

1. Traceability: the agent’s activities must be auditable, and:

2. Governance: the agent must follow my organization’s internal policies.

Additionally, I’d recommend designing agents with a human in the loop at critical decision points where someone with responsibility over an outcome can review the agent’s work and either approve it or order a revision.

The other side is culture, which I could write another whole article about.

Leadership preparing to make their workforce AI-enabled is going to see the best outcomes if their people feel empowered to up-skill and experiment with AI tools in their day-to-day work. This means genuinely buying into the idea that AI is about creating superpowers, not wholesale replacement. It’s up to leadership to understand this and trace out their AI enablement roadmap and share it with the organization.

If you’re going about this the right way, folks will hopefully see how collaborating with agents will make their work lives better. But so much depends on trust and transparency.

5. Remember: AI is still just software

Just because AI agents are cool doesn't mean they're magic.

I've seen leadership fall into this trap a lot, too, where expensive new technology is rolled out without the foundational work that makes any new software successful.

The principles of people-process-technology in balance still apply.

You still need:

1. Clear strategic vision

2. Defined success metrics

3. A well-managed process.

This is where you as Chief of Staff can have a huge impact as the bridge between functional leaders, technical implementation, and the rest of the org.

You're uniquely positioned to spot the gaps between what the technology promises and what your organization actually needs, then build the processes and partnerships to close those gaps.

At the end of the day, AI agents are just another way of getting work done, and their success depends on the same strategic thinking and cross-functional execution that you bring to everything else.

---

If you’re thinking about bringing AI agents into your organization, I’d love to chat about it. Connect with me on LinkedIn, reach out at sawyer@aomni.com or reach out here.