Context Switching Is Metabolic Interval Training

If you’re using AI well, you’re shipping a week of work in a day. That’s not a flex. It’s the new baseline, and it comes with a tax that nobody warned you about.

At any given moment during my workday, I have 8-10 different Claude Code agent sessions running across work codebases. Some of those sessions are managing teams of sub-agents within themselves, coordinating parallel workstreams that I need to review, redirect, or unblock. On top of that, I’m running 4-5 personal Claude Code instances working on completely separate problems. Fifteen sessions, fifteen mental models, fifteen different codebases with different constraints and different state.

I’m not writing the code anymore. I’m managing the harness, which means crafting the inputs, evaluating the outputs, reviewing the PRs the agents open, approving diffs, triaging test failures, reading logs, approving deployments, and tracking the lifecycle of every change from intent to production. The agents handle the keystrokes. I handle every decision around them, simultaneously, all day.

The intuition that AI makes work easier is wrong in a specific way. AI makes the implementation faster, which means the decisions compound. When you can ship 7x more code per day, you make 7x more architectural choices, evaluate 7x more agent outputs, review 7x more PRs, and rebuild your mental model 7x more often. The bottleneck moved from your fingers to your prefrontal cortex.

Gloria Mark’s research at UC Irvine produced the widely-cited finding that it takes about 23 minutes to fully regain focus after a significant interruption. Iqbal and Horvitz’s Microsoft Research field study found that 27% of alert-triggered task suspensions resulted in more than two hours before users returned to the original work. Rubinstein, Meyer, and Evans showed that task switching can consume up to 40% of productive time and reroutes neural firing through the prefrontal cortex’s executive-control network, increasing metabolic demand on a region that is already energy-expensive.

That’s the cost of one switch. Now multiply by fifteen sessions, all day, every day. That’s not a willpower problem. It’s a metabolic one.

A 2022 study by Wiehler et al. published in Current Biology used magnetic resonance spectroscopy to monitor brain chemistry across an entire workday. Participants doing cognitively demanding tasks accumulated significantly higher glutamate in the lateral prefrontal cortex. That glutamate buildup makes PFC activation progressively more metabolically expensive throughout the day. The fatigue you feel at 3 PM isn’t psychological. It’s a biological protective signal.

A March 2026 Harvard Business Review report coined the term “AI brain fry” for mental fatigue from sustained AI tool oversight. Workers reported a buzzing sensation, mental fog, difficulty focusing, and slower decision-making. The report identified a diminishing returns threshold where adding more AI tools actually reduced productivity gains. AI-assisted development shifts cognitive load from implementation (which the agents handle) to architecture-level abstract reasoning (which taxes the PFC harder), and you’re running that cycle across every parallel session with no recovery between rounds. The agents don’t get tired. You do.

Creatine Cognitive Performance: What the MRS Data Shows

The phosphocreatine system regenerates ATP at speeds the slower metabolic pathways cannot match. That speed is the entire reason it matters for cognitive demand spikes.

If the phosphocreatine shuttle is the bottleneck during cognitive demand spikes, the obvious question is whether supplementing creatine expands that buffer.

The answer from MRS (magnetic resonance spectroscopy) brain imaging: yes, oral creatine supplementation does increase brain creatine levels. But the kinetics are completely different from muscle.

Parameter	Muscle	Brain
Loading time	5-7 days	4-8 weeks
Typical increase	~20%	~8-11%
Transporter saturation	Not a constraint	Near saturation at the blood-brain barrier
Response to higher doses	Diminishing returns	Dose-dependent up to 10g/day

The blood-brain barrier creates a genuine bottleneck. Creatine crosses exclusively via the SLC6A8 transporter, which operates near saturation under normal conditions and actually downregulates with prolonged supplementation. The brain also compensates by reducing endogenous creatine synthesis, partially offsetting exogenous intake.

This means the standard gym-bro dose of 5g/day, which saturates muscle in a week, produces a much smaller and slower brain response than higher doses. Kondo et al.’s dose-response MRS data tells the story: 4.1% frontal PCr increase at 4g/day versus 9.1% at 10g/day. The brain’s dose-response curve hasn’t plateaued at standard doses. It roughly doubles between 4g and 10g (small sample, female adolescents with depression, so generalizability is limited but the dose-response shape is informative).

The Evidence: Honest Version

Here’s where I need to be straight with you about creatine cognitive performance claims, because the supplement industry won’t be.

A 2024 meta-analysis of 16 RCTs (492 participants) found statistically significant effects for memory (SMD=0.31, p<0.00001), attention time (SMD=-0.31, p=0.03), and processing speed (SMD=-0.51, p=0.04). But a 2026 commentary identified a unit-of-analysis error in that meta-analysis that may inflate the results.

The largest, best-designed RCT (Sandkühler et al. 2023, N=123, published in BMC Medicine) found effects of d=0.09 for abstract reasoning and d=0.17 for working memory. In IQ terms, that’s roughly 1 and 2.5 points. Detectable in large samples, not something you’d notice subjectively.

The European Food Safety Authority formally rejected the cognitive health claim in November 2024, concluding that a cause-and-effect relationship had not been established. Researcher Darren Candow, who studies creatine, called the rejection “justified” “based on the current body of research.”

So why am I still interested?

Because the effects are not uniform across conditions. They concentrate heavily under metabolic stress.

The Gordji-Nejad 2024 study gave participants a single high dose of creatine (0.35g/kg, roughly 25g for a 70kg person) during 21 hours of sleep deprivation. MRS brain scanning confirmed increased phosphocreatine, maintained ATP levels, and prevented pH decline. Processing speed improved 16-29%. Memory improved 10%. Effects peaked at 4 hours and lasted 9 hours. The creatine group maintained near-normal brain energy levels while the placebo group’s depleted measurably.

McMorris et al. (2006) found that 20g/day for 7 days significantly preserved cognitive performance during 24-hour sleep deprivation, with a “positive effect on tasks that place heavy stress on the prefrontal cortex.”

Older adults (66-76) showed a large memory effect (SMD=0.88, p=0.009). Younger adults in the same analysis showed essentially zero (SMD=0.03).

The pattern: creatine helps when the brain’s energy system is already under pressure. Rested, well-fed, single-task conditions show minimal benefit. Sleep deprivation, sustained cognitive fatigue, aging, vegetarian diets (lower baseline creatine) show meaningful effects.

Eight hours of orchestrating fifteen agent sessions, each demanding a full mental model load, every PR review, every deployment decision, every architectural call across multiple codebases, is not a normal resting condition. It is exactly the kind of sustained metabolic stress where the clinical data shows creatine’s biggest effects. Whether that stress crosses the threshold where supplementation produces noticeable benefit is an open question, and I want to test the hypothesis rather than trust the marketing.

Dosing for the Brain

If you’re convinced enough to try it, here’s what the MRS data supports:

Standard (conservative): 10g/day, split into two 5g doses with meals. The Kondo dose-response data shows this is the minimum where brain phosphocreatine increase roughly doubles versus lower doses. Expect 4-8 weeks for meaningful brain saturation.

Aggressive (matches the strongest study protocols): 20g/day (4x5g with meals) for 4 weeks as a loading phase, then 10g/day maintenance. Dechent et al. achieved 8.7% brain creatine increase at this protocol over 4 weeks.

Body-weight adjusted (2024 review): Loading at 0.3g/kg/day, maintenance at 0.10-0.14g/kg/day. For an 80kg person, that’s 24g loading, 8-11g maintenance.

Form: Micronized creatine monohydrate. Nearly every cognitive study used monohydrate. Two cognitive trials exist on other forms (Ling et al. 2009 ethyl ester, CONCRET-MENOPA 2025 HCl), but monohydrate is by far the most-studied and the cheapest. Buffered forms and gummies still have no published cognitive data. The ingredient list should contain one item.

Timing: Doesn’t matter. Creatine saturates stores over weeks. Once saturated, the phosphocreatine system is available 24/7. Morning with breakfast is practical because you pair it with carbs (insulin helps cellular uptake) and build a routine.

Cycling: Unnecessary. No tolerance develops. The ISSN position stand documents safety up to 30g/day for 5 years with no adverse effects.

Cost: Creapure-certified creatine monohydrate runs roughly $40-60 per kilogram across major retailers. At 10g/day that works out to $12-18/month and lasts about 100 days per kilogram. Creapure certification means 99.9%+ purity, manufactured in Germany, third-party tested.

Safety

I went deep on the safety data because “just take more” is not a serious recommendation without it. Three findings stand out.

Kidney function is fine. A 2025 BMC Nephrology meta-analysis found no significant GFR changes. A 2025 Frontiers in Nutrition safety review reported “no evidence of renal function impairments” across protocols ranging from 1-80g/day for 5 days to 60 months (most trials used standard 3-25g/day; 80g/day is the upper extreme).

Hair loss is unlikely. The fear traces to one 2009 study of 20 rugby players where DHT rose 56% after loading and stayed 40% above baseline during maintenance, though all absolute values stayed within the normal reference range and hair was never measured. Twelve subsequent trials measured testosterone with no significant changes. The first study to directly measure hair follicles, Lak et al. 2025, found no DHT change and no hair density or thickness change after 12 weeks.

Tell your doctor before blood work. This is the one that catches people. Creatine supplementation elevates serum creatinine, the standard blood marker for kidney function. Published cases exist where creatine users were falsely diagnosed with acute kidney failure based on creatinine alone. The National Kidney Foundation explicitly lists creatine as a factor that artificially raises creatinine. If your annual physical shows elevated creatinine, request a cystatin C test for accurate kidney assessment.

GI distress is real at high single doses. One study found 55.6% diarrhea incidence at a single 10g dose versus 28.6% for split 5g doses. Always split to 5g or less per serving. Take with food.

If you’re shipping a week of work in a day, the bottleneck is no longer the keyboard. It is the wet hardware between your ears.