On New Year's Day 2026, programmer Steve Yegge launched Gas Town, an open-source platform that lets users orchestrate swarms of AI coding agents simultaneously, assembling software at speeds no individual developer could match.
While the results were impressive, early users also said witnessing it was overwhelming. One described a "palpable sense of stress" watching the system work.
Gas Town was moving too fast for the humans operating it.
That reaction should concern every marketing leader currently scaling AI across their team.
A January 2026 BCG study of 1,488 full-time U.S. workers has put numbers on something many marketing leaders have felt forming in their teams but lacked language for. The researchers call it "AI brain fry," which they define as mental fatigue from excessive use or oversight of AI tools beyond one's cognitive capacity. Participants described a buzzing sensation, mental fog, difficulty focusing, slower decision-making, and headaches severe enough to require physically stepping away from their computers1.
Of all the functions surveyed, marketing professionals reported the highest prevalence. Twenty-six percent of marketing workers endorsed the experience of AI brain fry, ahead of people operations at 19%, operations and engineering at 18%, and finance at 17%.
The finding that would keep me up at night, though, is this: among workers who did not report brain fry, 25% showed active intent to leave their organization. Among those who did report it, that number rose to 34%. A 39% increase in attrition intent, concentrated in the employees using AI most intensively. The people most organizations consider their highest performers.
The problem is not that AI tools don't work. The problem is how organizations are designing AI into the work.
The Cognitive Cost Nobody Budgeted For
AI adoption is no longer an initiative. It is a baseline expectation. Eighty percent of employees now use AI tools at work, a 52% increase from two years ago2. In parallel, 53% of leaders say productivity must increase further still3.
The assumption embedded in most enterprise AI strategies is that more adoption produces proportionally more output. The BCG data dismantles that assumption with unusual precision.
As employees move from one AI tool to two used simultaneously, self-reported productivity increases significantly. Adding a third tool produces another gain, though smaller. After three tools, productivity scores decline4. The relationship is not linear. It is a curve with a peak, and most enterprise AI strategies are designed to push workers past it.
The cognitive costs accumulate on the other side of that peak. Workers whose AI tools require high degrees of direct oversight expend 14% more mental effort and experience 12% more mental fatigue than those with low-oversight AI use. They report 19% greater information overload, the experience of being overwhelmed by the sheer volume of inputs they must process, evaluate, and act on in the same number of hours5.
These findings do not exist in isolation. ActivTrak's analysis of 443 million working hours finds that focus efficiency has dropped to 60%, a three-year low, even as collaboration time surged 34% and multitasking rose 12%6. Microsoft's 2025 Work Trend Index reports that 80% of the global workforce says they lack sufficient time or energy to do their work7. Gartner now lists "the mental toll of near-ubiquitous GenAI adoption" as a trend requiring urgent organizational attention in 20268.
The pattern across these independent sources is consistent. Organizations are layering AI tools onto existing workloads without redesigning the work itself. The cognitive bill is coming due. And it is arriving first at the desks of the people organizations can least afford to lose.
Workers who report AI brain fry experience 33% more decision fatigue. They report making minor errors 11% more frequently and major errors (those affecting safety, outcomes, or important decisions) 39% more frequently. They are, as the BCG researchers document, more likely to be actively looking for the exit.
A separate HBR study from February 2026 mapped the same dynamic from the other direction. Researchers found that AI did not reduce work for the employees studied. It intensified it. Workers moved faster, took on broader scope, and extended their hours. Often without being asked9. The tools amplified capacity and the organization absorbed every drop of it, then asked for more.
This is not a people problem. It is a design problem. And naming it correctly matters, because the default enterprise response, more training, more AI champions, more enablement, treats the design problem as a skills gap and makes it worse.
The Tradeoff Most CMOs Don't See Until It's Too Late
The default enterprise AI strategy looks like this:
- Deploy AI tools across the team.
- Measure adoption rates.
- Track productivity gains.
The assumption is that more adoption = more output = better ROI.
But the data shows a different pattern:
You're trading short-term productivity gains for long-term attrition risk.
BCG found that 34% of workers who report AI brain fry are actively looking to leave. That is a 39% increase in attrition intent compared to workers who don't report brain fry.
Let's do the math on what that means for a marketing team of 50 people:
- If 26% of marketing professionals report brain fry (BCG's finding), that's 13 people on your team.
- If 34% of those 13 people are actively job-hunting, that's 4 people looking for the exit.
- And it’s not just any 4 people, your highest performers are often the ones using AI most intensively.
The cost of replacing a single marketing professional is $75,000–$150,000 (recruiting, onboarding, ramp time, lost productivity).
So the hidden cost of asking your team to orchestrate AI themselves, instead of embedding a specialist who handles that cognitive load, is $300K–$600K in attrition risk.
And that's before you factor in the productivity loss from the decision fatigue, errors, and focus fragmentation the BCG study documents.
The specialist model flips that tradeoff:
- You invest $100K–$150K in an embedded AI-fluent specialist for 6–12 months.
- They handle the orchestration burden, reducing mental strain by 15% (per BCG).
- Your team captures the productivity gains without the cognitive overload that's driving your best people toward the exit.
- And when the specialist rolls off, your team retains the orchestration capability because they developed it through proximity, not through training programs.
The ROI goes beyond just the improved workflows the specialists redesigns for your team, it’s also in the people you don't lose.
Trying to train everyone to become an AI expert is a clear path to burnout.
Conversely, embedding a specialist who already is an expert and who rebuilds workflows that make life easier for everyone else? That is a true structural fix.
You're Asking Every Pilot to Be Their Own Air Traffic Controller
Consider how commercial aviation handles complexity.
A modern air traffic controller manages between 10 and 20 aircraft simultaneously, monitoring trajectories, sequencing approaches, and coordinating with pilots who each see only their own flight path. The cognitive demands are extraordinary: sustained attention across multiple dynamic systems, rapid switching between communication channels, continuous risk assessment under time pressure.
Aviation did not solve this problem by training pilots to manage each other's flight paths. It recognized that orchestration is a fundamentally different function from execution. The cognitive work of monitoring, coordinating, and prioritizing across multiple simultaneous systems requires dedicated capacity. Pilots fly aircraft. Controllers manage the airspace. The division exists because asking one person to do both degrades performance at both tasks.
Enterprise AI strategy has arrived at the equivalent of asking every pilot to also manage the airspace around them.
Marketing teams now operate across content generation agents, data analysis tools, campaign automation platforms, research assistants, image generators, and coding copilots. Each tool requires oversight. Each produces outputs that must be evaluated, corrected, and integrated into workflows designed for human-only execution. The cognitive task is not using any single tool. It is orchestrating the interplay between all of them while still doing the strategic, creative, and relational work that constitutes the actual job.
The BCG data makes the consequences precise. Brain fry doesn't stem from AI use itself. It stems from the oversight intensity. When organizations measure and reward AI tool consumption, as Meta does by including lines of AI-generated code as a performance metric for engineers, they incentivize exactly the behavior pattern that produces cognitive overload.
Cognitive load theory explains the mechanism. Human working memory is sharply limited in capacity. When task demands exceed that capacity, performance degrades and mental fatigue accumulates, not gradually but at an accelerating rate once the threshold is crossed10. Each additional AI agent requiring oversight doesn't add a proportional increment of cognitive load. It compounds the switching, monitoring, and integration burden on a system that was already operating near capacity.
Sophie Leroy's research on attention residue quantifies one dimension of this cost. When a person switches from Task A to Task B, cognitive residue from Task A persists, reducing performance on Task B. The effect is not momentary. It lingers11. Now multiply that switching cost across four, five, six AI tools, each generating outputs that require evaluation before the next task can proceed. The compound cognitive tax explains why productivity declines after three simultaneous tools despite each individual tool delivering genuine value in isolation.
The air traffic control parallel is not decorative. It is structural. Aviation solved this exact problem by concentrating the orchestration function in dedicated specialists whose cognitive capacity is professionally calibrated for multi-system oversight. The question for enterprise AI strategy is whether organizations will arrive at the same structural insight, or continue distributing the orchestration burden across people whose primary job, and primary cognitive capacity, is needed elsewhere.
What Aviation Understood That Enterprise AI Strategy Hasn't
The BCG study contains a finding that points directly toward the structural fix.
When teams have organized integration of AI into their processes, when AI is embedded in the workflow as a collective capability rather than layered onto individuals as an additional responsibility, team members experience significantly less mental strain. Workers whose managers actively answer their questions about AI report 15% lower mental fatigue scores. And when AI is used to replace routine, repetitive tasks rather than to intensify oversight demands, burnout scores drop 15%12.
These findings describe the embedded specialist model before giving it a name.
An AI-fluent specialist embedded inside a marketing team concentrates the orchestration burden in someone whose cognitive capacity is built for it. They handle the multi-tool coordination, the agent oversight, the output evaluation, and the workflow integration that produces brain fry when distributed across the team.
The team's relationship to AI shifts from high-oversight (the brain fry zone) to organized integration (the reduced-strain zone the BCG data identifies).
But something more valuable happens simultaneously: the orchestration skill transfers through proximity.
This is the mechanism that training programs cannot replicate. When a specialist sits inside the team and manages AI complexity as part of the daily work, team members observe how they prioritize which outputs to review, when to trust an AI-generated result and when to challenge it, how to sequence multi-tool workflows without fragmenting their attention. The learning is contextual and continuous. It happens not in a classroom but during the actual campaign build, the actual reporting cycle, the actual content production pipeline.
McKinsey's research supports the principle. "The real productivity unlock comes from reimagining workflows so people, agents, and robots each do what they do best"13. Half of organizations identified as AI high performers are already pursuing workflow redesign as their primary strategy14. They have recognized that the value of AI is not in how many tools each person uses. It is in how the work is designed so that human cognitive capacity is spent on judgment, strategy, and creative problem-solving rather than on monitoring agent outputs.
The practical picture: an embedded AI-fluent specialist arrives, audits the team's workflows, identifies where AI oversight is consuming cognitive capacity that should be directed at higher-value work, and rebuilds those workflows so AI handles the routine tasks (reducing burnout) while the specialist carries the orchestration load (preventing brain fry). As the team works alongside the specialist day after day, they absorb the orchestration instincts. Not because someone taught them in a workshop, but because they watched, collaborated, and gradually developed their own capacity for managing AI at sustainable cognitive levels.
The specialist's success is measured not by how long they stay. It is measured by the team's ability to manage AI independently after they leave, without the cognitive overload that was burning out their best people.
Here’s How Some Organizations Are Already Solving This
A small but growing number of marketing leaders have figured this out. Instead of asking their entire team to "get better at AI", they embedded a specialist inside the team.
That specialist came in with one specific purpose, to redesign workflows so the team could capture AI's productivity gains without the cognitive overload.
They handled the orchestration burden so the rest of the team could focus on strategy, creative, and execution while the specialist did the work to integrate AI into the team's existing processes
And here's what happened:
- Email production cycles that used to take days now take hours.
- Reporting processes that used to be weekly slogs are now automated.
- Campaign operations that used to require proportional headcount now scale across markets without adding bodies.
But the most important outcome wasn't the workflows, it was what happened after the specialist left:
The team retained the orchestration capability.
Most of the content learned in training programs don’t get applied and are lost within weeks. But when people develop capability through proximity by working alongside someone who demonstrated it daily, it sticks.
This is the embedded specialist model, and it's the structural fix the data points toward.
The Strategic Conversation This Data Makes Possible
Most marketing leaders are currently trapped in a narrative that sounds like this: "We invested in AI tools. We trained the team. Adoption is increasing. But something isn't working and I can't quite explain why."
The BCG research replaces that ambiguity with a specific, defensible diagnosis. The tools are working. The adoption is real. The problem is that distributing AI orchestration across the entire team is producing cognitive overload that erodes the productivity gains, increases errors, and drives the most capable people toward the exit. This is not a failure of execution. It is a predictable consequence of a specific design choice, one that can be redesigned.
That reframing changes the board conversation materially.
"Our team is struggling with AI" sounds like a capability problem. It invites responses like "invest in more training" or "hire people with more AI experience," neither of which addresses the structural issue. "Our AI strategy distributes the orchestration burden in a way that exceeds our team's cognitive capacity, and the data shows this is producing decision fatigue, errors, and attrition risk in our highest-performing people" sounds like a systems problem with a systems solution. It is the kind of language that positions the person saying it as someone who understands organizational architecture, not just technology adoption.
Thirty-two percent of leaders already plan to hire AI agent specialists within the next 12 to 18 months14. The market is moving toward the embedded specialist model because the data is making the alternative increasingly difficult to defend.
The organizations that embed AI-fluent specialists into their marketing teams now will see concrete outcomes: workflows that run at AI-enabled speed without the cognitive tax on the team. Email production cycles that compress from days to hours. Reporting processes that collapse from weekly slogs to automated outputs. Campaign operations that scale across markets without requiring proportional headcount. And crucially, the team retains the orchestration capability after the specialist departs because they developed it through months of working alongside someone who demonstrated it daily.
The BCG data shows that employees who feel their organization values work-life balance report 28% lower mental fatigue scores. When organizations expect employees to simply accomplish more work because of AI, mental fatigue scores rise 12%.
The signal an organization sends about AI's role matters as much as the tools themselves. Embedding a specialist sends a specific signal: we are investing in making AI work for the team, not asking the team to sacrifice their cognitive health for AI adoption metrics.
The Specialist Role That Enterprise AI Is Missing
Commercial aviation did not make flying safer by training pilots harder. It recognized that orchestration is a specialist function, that the cognitive work of managing multiple simultaneous systems requires dedicated capacity, not distributed effort across people whose primary expertise and attention belongs elsewhere.
Enterprise AI strategy is at that recognition point now.
The organizations that act on it will retain the employees who are currently burning out under the orchestration burden. The 26% of marketing professionals already reporting brain fry, and the larger population approaching the threshold. They will capture the productivity gains that AI genuinely delivers when its integration is designed rather than improvised. And they will build teams that develop AI orchestration capability sustainably, through proximity to specialists who carry that cognitive load as a professional function, not through training programs that ask people to learn a new cognitive discipline in the 24 minutes per week they have available for formal learning15.
The air traffic controller doesn't make the pilot less important. They make the pilot more effective by handling the cognitive work the pilot shouldn't be doing while flying the plane.
Your marketing team is flying the plane. The question is whether you are going to ask them to manage the airspace too, or whether you are going to put someone in the tower.
What the Embedded Specialist Model Looks Like in Practice
Here is a brief look at how an embedded AI-fluent specialist would transform an enterprise marketing team within 6–12 months.
Month 1: They audit workflows, mapping where AI oversight is consuming cognitive capacity that should be directed at higher-value work.
Months 2–4: They rebuild those workflows so AI handles routine tasks, the team focuses on judgment, and the specialist handles orchestration.
Example: A 3-day campaign reporting process becomes a 3-hour process.
Months 5–8: The team absorbs orchestration instincts through proximity—learning how to structure prompts, sequence multi-tool workflows, and validate outputs without burning out.
Months 9–12: The specialist rolls off. The team is left with workflows that run at AI-enabled speed, the orchestration capability to manage AI independently, and the documentation to onboard new hires into the system.
The specialist's success isn't measured by how long they stay.
It's measured by the team's ability to manage AI independently after they leave, without the cognitive overload that was burning out their best people.
Be on the lookout for an upcoming full guide on how to embed an AI specialist inside your marketing team.
The air traffic controller doesn't make the pilot less important. They make the pilot more effective by handling the cognitive work the pilot shouldn't be doing while flying the plane.
Your marketing team is flying the plane.
The question is whether you're going to ask them to manage the airspace too, or whether you're going to put someone in the tower.
If you want to understand where the orchestration burden sits in your team, and what a structural fix would involve, Algomarketing can show you starting in our first conversation.
Want to see where your AI strategy is distributing cognitive load in a way that’s burning out your best people?
