What the Gartner AI Agent Forecast Means for Ops Leaders — Before Your Q3 Planning Window Closes

Two weeks ago, my CEO forwarded me a newsletter. The subject line of his email was one word: "Thoughts?"

The newsletter contained the Gartner forecast — the one predicting that 40% of enterprise applications will embed task-specific AI agents by end of 2026, up from less than 5% in 2025. I had roughly 48 hours before it came up in our weekly 1:1. I needed a useful answer, not an interesting observation.

I spent the first hour reading the coverage. Fourteen browser tabs. Every single one was written for someone evaluating software platforms — not for someone who has to decide, before Q3 planning locks in, what her operations function actually looks like with AI agents running parts of it. I closed all fourteen tabs and started from a different question entirely.

The Short Answer

For COOs and Chiefs of Staff, the Gartner forecast means one thing operationally: the question is no longer whether AI agents will handle parts of your business, but which workflows to delegate first. The right entry point is not the most impressive AI capability — it is the highest-volume, most-repetitive, lowest-stakes work your team does every week. Identifying those workflows takes 30 minutes. Every ops leader should complete that exercise before their Q3 planning cycle closes.

Why Every Coverage Piece on This Forecast Is Written for the Wrong Person

I do not need to know which AI agent platform has the best workflow orchestration engine. I need to know which of my current workflows — the ones consuming my team's hours every week — should have been automated six months ago. Those are two different starting points, and they produce two different Q3 plans.

I know I am not the only ops leader staring at this gap. A G2 report from 2025 found that 57% of companies already have AI agents in production. More than half plan to expand scope or increase budgets in the next twelve months. The adoption is real. But when I went looking for guidance on how to adopt — specifically for someone who runs the business, not someone who runs the infrastructure — I found almost nothing. Platform comparisons, architecture diagrams, vendor rankings. Zero help with the question I actually had: which of my workflows do I hand over first, and what do I keep?

That is where most operations teams are stuck right now. Not stuck on whether AI is real. Stuck on where to start.

The Right Question: Which Workflows Should I Have Automated Six Months Ago?

When I stopped reading platform comparisons and started looking at my own team's week, the answer became obvious within thirty minutes. I pulled up our operational calendar and asked three questions about every recurring workflow:

How often do we do this? If the answer is more than weekly, it stays on the list.

How similar is each instance? If the process is substantially the same every time — same inputs, same structure, same output format — it stays on the list.

What happens if the AI gets it wrong? If a human can catch the error before it creates a real problem, it stays on the list. If a mistake goes directly to the board or a customer with no review step, it comes off.

Three criteria. Thirty minutes. The workflows that scored highest on all three were not surprising: status update collection across teams, check-in aggregation before our Monday leadership meeting, report data-gathering for monthly reviews, and meeting prep summaries for our CEO's weekly schedule. Every one of these consumed hours of senior human time. Every one followed the same pattern every cycle. And every one had a natural human review point built in.

The workflows that did not pass the filter were equally clear: strategic decision support, cross-functional conflict resolution, performance management conversations, and anything requiring an understanding of the political dynamics behind the data. Those still need a person — specifically, they need the judgment that a Chief of Staff or VP of Ops brings to the table.

The exercise took half an hour. I walked into my CEO's 1:1 with a prioritized list of four workflows and a timeline for each. That was the answer to "Thoughts?" — not a position paper on the future of agentic AI, but a plan for what changes in our operations function next quarter.

AI-Assisted vs. AI-Run: The Distinction That Actually Matters

Here is the distinction I wish someone had made for me a year ago, because it would have saved me three months of miscalibrated expectations.

AI-assisted means the AI helps me do something faster. It drafts a status update. It suggests a priority. It flags a data point I might have missed. I still do the work. The AI makes me more efficient at it.

AI-run means the AI completes the entire workflow. It gathers data from Jira, Salesforce, and HubSpot. It generates the status report. It routes the report to the right people. It triggers the next step in the cadence. I show up only at the decision points — the moments where judgment matters.

The difference is not incremental. It is structural. AI-assisted shaves minutes off tasks. AI-run eliminates entire workflow categories from my team's calendar.

Most of the AI content written for operations leaders conflates these two modes. They describe AI-assisted improvements with the language of transformation. The result is that ops leaders expect transformation and experience marginal efficiency gains — which feels like a failure even when the tool is working exactly as designed.

The governance requirements are completely different too. AI-assisted is low-risk — if the AI drafts a bad status update, I catch it and rewrite it. AI-run requires designing the oversight architecture deliberately. For our Monday leadership prep, I decided the AI owns data gathering, formatting, and delivery. It escalates when a metric deviates more than 15% from the prior week or when a data source fails to connect. Everything else ships without my involvement. That design conversation took an afternoon, and it saved my team four hours every single week.

The first time I watched an AI agent run our Monday leadership prep — pull live data from six connected tools, compile it into the format our CEO reads, flag the three items that needed human attention, and deliver it at 6am without anyone touching it — I did not feel automated. I felt like I finally had the leverage to spend Monday morning on the decisions rather than on the assembly work that preceded them.

The second workflow we moved to AI-run was our weekly cross-functional check-in aggregation. Every Friday, five department leads submitted updates through five different formats — Slack messages, email bullets, a shared doc, a Loom video, and one person who just called me. The AI agent now collects structured check-ins from each lead, normalizes the format, flags conflicting priorities between teams, and delivers a single synthesized brief by Thursday evening. I used to spend three hours every Friday chasing and compiling those inputs. Now I spend twenty minutes reviewing the brief and adding the context only I can see — who is sandbagging their forecast, which team is about to miss a dependency nobody else has noticed.

That is what AI-run looks like from the inside. Not replacement. Architectural leverage.

How to Build Your H2 AI Operations Roadmap in 30 Minutes

If you are a COO, Chief of Staff, or VP of Ops reading this before your Q3 planning cycle closes, here is the exercise I ran. It takes thirty minutes and produces a prioritized list you can bring to your next planning conversation.

Step 1: List your ten most frequent operational workflows. Not your most strategic. Your most frequent. The things your team does every week or every cycle without exception. Status collection. Report prep. Meeting summaries. Review data-gathering. Check-in follow-ups. Goal progress aggregation. Decision re-communication. Get them all on a list.

Step 2: Score each workflow on three dimensions. Volume (how often — daily, weekly, monthly). Repeatability (how similar is each instance, on a scale of 1 to 5). Risk (what happens if the AI gets it wrong — low, medium, high). Use a simple spreadsheet. Do not over-engineer the scoring.

Step 3: Identify your top three. The workflows with the highest volume, highest repeatability, and lowest risk are your H2 AI agent priorities. These are the workflows where the return is highest and the downside is most contained.

Step 4: For each of the top three, decide: AI-assisted or AI-run? If the workflow can be completed end-to-end with a human review at the output stage, it is a candidate for AI-run. If the workflow requires human judgment at multiple intermediate steps, start with AI-assisted and evolve.

Step 5: Put these three into your Q3 operating plan. Not as a vague "explore AI" initiative. As specific workflow transitions with owners, timelines, and success metrics. "Transition Monday leadership prep from manual to AI-run by end of July, with parallel running for the first four weeks" is a plan. "Evaluate AI agent platforms for potential operational improvements" is a committee, and committees do not produce operational change.

Three workflows. Not a company-wide AI strategy. Not a platform evaluation. Three workflows, prioritized by the math, with clear owners and a defined timeline. I have watched two peer companies spend all of Q2 in "AI evaluation mode" — running pilots with no decision criteria, comparing platforms with no workflow targets. They will enter Q3 exactly where they started. Meanwhile, up to 70% of their management time continues to disappear into coordination work — status updates, report prep, data chasing — the exact category AI agents are built to handle. The cost of delay is not abstract. It is measured in senior hours that never come back.

The Reflection

I keep that "Thoughts?" email in my inbox. Not as a reminder of the Gartner forecast — I have read enough Gartner coverage for a lifetime. I keep it because of what it taught me about my own role.

For three years, I thought being indispensable to every operational workflow was the point. Being the person who could pull the numbers, compile the brief, route the decision, chase the follow-up — that was the job. When AI agents started handling those workflows, I did not lose my job. I discovered what my job was supposed to be. The judgment calls. The pattern recognition across six functions that no dashboard captures. The conversation with the CFO where the real question is three layers beneath the question she actually asked.

The Gartner forecast is not about platforms. It is about what happens to your role when the coordination layer stops requiring you to operate it by hand. For some ops leaders, that is a threat. For the ones who were always constrained by the assembly work rather than defined by it, it is the first time the role matches the potential.

If you're the person who got the "Thoughts?" email and you still don't have an answer, try Rhythms for free at rhythms.ai and see what it looks like when the coordination layer runs itself.

Frequently Asked Questions

How are COOs actually using AI agents in their operations in 2026?

The most common adoption pattern among ops leaders I talk to is starting with the coordination layer — the workflows that exist purely to enable other workflows. Status collection, report data-gathering, check-in aggregation, and meeting prep are the four entry points I see most frequently. These are high-volume, highly repetitive, and low-risk when the AI gets something slightly wrong. The COOs who are furthest ahead started with these four and expanded from there, rather than starting with a company-wide AI strategy.

What workflows should a Chief of Staff delegate to AI agents first?

Apply the three-criteria filter: high volume (you do it more than weekly), high repeatability (the process is substantially the same each time), and low catastrophe risk (if the AI gets it wrong, a human catches it before it reaches anyone external). Status update collection, check-in aggregation, report data-gathering, and meeting prep all pass this filter. Strategic decision support, performance management, and cross-functional conflict resolution do not. Start with the first category. Leave the second to humans.

What is the difference between AI-assisted and AI-run operations?

AI-assisted means the AI helps you do something faster — it drafts, suggests, flags. You still do the work. AI-run means the AI completes the entire workflow end-to-end, with human oversight only at defined decision points. The distinction matters because the trust model, governance requirements, and operational impact are completely different. Most disappointment with AI in operations comes from expecting AI-run results from an AI-assisted deployment.

How do you build trust in AI-generated operational outputs?

Parallel running. Have the AI agent complete the workflow alongside your existing human process for four to six weeks. Compare outputs side by side. When the AI's outputs are consistently within acceptable variance of the human's, the human step becomes a review rather than a re-creation. The trust builds from the comparison, not from a decision to trust. Skip the parallel period and you will spend months second-guessing every output the system generates.

How do you decide which operations workflows are safe to automate with AI?

Score each workflow on three dimensions: volume (how often you do it), repeatability (how similar each instance is), and risk (what happens if the output is wrong). High volume, high repeatability, low risk — those are safe to automate now. Low volume, low repeatability, high risk — those stay with humans. The middle category is where judgment matters: start with AI-assisted for those workflows and graduate to AI-run as the system proves itself.