CO2, Bad Decisions, and the Case for AI Agents

Your team's worst decisions might not be a strategy problem. They might be an air quality problem.

A growing body of research shows that CO2 concentrations in typical meeting rooms, often 1,000 to 2,500 ppm, meaningfully degrade cognitive function. Complex decision-making takes the hardest hit. Mike Bowler's recent writeup on this landed with a lot of engineering and ops folks because it reframes a frustrating pattern: smart people, sitting together, making surprisingly bad calls.

That's worth sitting with before we talk about agents.

What the CO2 research actually says

Outdoor air runs around 420 ppm. A closed conference room with six people in it can hit 1,500 ppm within 45 minutes. Studies from Harvard and Lawrence Berkeley found that at 1,000 ppm, scores on complex decision-making tasks drop by roughly 15 percent. At 2,500 ppm, they drop by 50 percent.

The decisions that suffer most are the ones requiring working memory, strategic thinking, and weighing multiple variables, which is exactly the kind of reasoning that goes into scoping a project, evaluating a vendor, or designing a process.

The uncomfortable implication for how we run teams

Most companies still structure their highest-stakes decisions as synchronous, in-room events: planning meetings, sprint reviews, architecture discussions, budget calls. The medium we've chosen for complex reasoning is one of the worst possible environments for it.

This isn't about getting a CO2 monitor (though that's not a bad idea). It's about recognizing that human cognition is a variable, not a constant. It degrades under load, fatigue, and yes, ambient air quality. Building processes that treat human judgment as perfectly reliable is a design flaw.

Where AI agents actually fit into this

Agents aren't here to replace judgment. But they're very good at handling the parts of a workflow that shouldn't require judgment in the first place.

The tasks that burn through your team's cognitive budget without needing to: routing incoming requests, summarizing context before a decision, pulling the relevant data before a meeting, drafting the first version of a document, following up on open threads. These are the tasks that fill the day before anyone gets to the thing that actually matters.

When an agent handles the intake, the research, the first draft, and the follow-through, the human in the loop arrives at the decision point with more working memory to spend on it, not less.

There's a framing that helps here: AI agents work best as decision-support infrastructure, not decision-makers. The goal isn't to automate the judgment. It's to protect the conditions under which good judgment is possible.

How to spot where your team is bleeding cognitive load

Most teams have a handful of workflows that are quietly expensive in this way. They're usually characterized by:

• High frequency (happens multiple times a week)
• Structured inputs (the same information is gathered every time)
• A human in the loop whose job is mostly to read and forward, not actually decide
• Delays caused by waiting, not by complexity

If a process fits that description, it's a candidate for an agent that handles the structured work and hands off only the genuine decision points.

Our free AI Opportunity Audit scans your website and identifies your three highest-impact automation candidates in a few minutes. It won't tell you to automate everything, but it will tell you where the low-hanging friction is.

The meeting room is a clue, not the problem

Bowler's article is really about noticing invisible constraints. CO2 is one. But the broader pattern is that organizations consistently underestimate how much environmental and logistical overhead degrades the quality of the work they care most about.

Building AI agents into your workflows is, in part, a bet that your team's attention and judgment are worth protecting. The automation isn't the point. The headroom it creates is.

Want this kind of automation built for your team?

If this is the kind of infrastructure you want in place, we build it. Book a call and we can talk through where it fits your specific workflows.