Sign in Subscribe

AI: Friend or Foe for Flow?

AI: Friend or Foe for Flow?
Photo by Steve Johnson / Unsplash

Why the question isn't whether to use AI. It's whether you're using it in a way that protects or dismantles your most valuable cognitive state.

Estimated read time: 15+ minutes. In a hurry? Skip to the TL;DR.

Grab a coffee (or wine) and settle in. This is a big one and it is worth reading properly (if reading on email it may get clipped due to length, read the full article online.

Here's a question I've been sitting with for a few months.

If AI is supposed to make us more productive, why are so many high-performing leaders I work with reporting that their work feels less satisfying, less creative, and somehow less theirs, even as they're technically producing more output, faster?

The answer, I think, lies in what we're actually optimising for.

Most people are using AI to optimise for throughput: how much gets done, how quickly. What they're unintentionally sacrificing is depth. The quality of cognitive engagement that produces their best thinking and, not coincidentally, makes work feel meaningful.

The neuroscience term for that depth is flow. And AI, depending on how you're using it, is either the most powerful flow-amplifier available to you right now, or a sophisticated machine for preventing you from ever reaching it.

This edition is about knowing the difference. And using AI with intention.

A Note on the Research

I want to be upfront: direct peer-reviewed research specifically studying AI and flow states is essentially non-existent at this point. We're too early in the adoption curve for that kind of rigorous longitudinal work.

What does exist is a substantial body of converging evidence: decades of flow neuroscience, strong research on technology and attention, well-established cognitive offloading literature, and several impressive recent field experiments on generative AI and knowledge worker productivity. This edition synthesises all of it.

Where I'm making inferential leaps rather than citing direct evidence, I'll tell you.

Flow: The State Your Best Work Comes From

If you've been reading this newsletter for a while, you'll know I come back to flow often. It's the neurological state I build most of my coaching work around, and the one I'd argue represents the highest-leverage intervention available to senior leaders.

But let's make sure we're on the same page about what it actually is, because the word gets misused constantly.

Flow is a state of optimal experience characterised by complete absorption in a challenging task, loss of self-consciousness, distorted time perception, and deep intrinsic reward. Csikszentmihalyi first described it formally in 1990. Kotler and the Flow Research Collective have spent the past two decades mapping the neuroscience and performance data around it.

The performance numbers are striking. McKinsey's research on senior executives found those in flow report being approximately five times more productive than in their normal working state. For my fellow geeks out there, this isn't a motivational claim. It reflects a coordinated neurochemical state involving dopamine, norepinephrine, endorphins, anandamide, and serotonin. A cocktail that simultaneously boosts motivation, attention, pattern recognition, creative association, and implicit learning (van der Linden, Tops, & Bakker, 2021).

Neurobiologically, flow involves transient hypofrontality: a temporary downregulation of the dorsolateral prefrontal cortex (DLPFC), the region responsible for self-monitoring, inner critic activity, and conscious self-regulation (Dietrich, 2004; Ulrich et al., 2014). Your inner critic goes quiet. Your task-relevant networks fire hard. You think faster and better than in normal conscious effort.

There are many flow triggers but the three core conditions are required to get there:

  1. Challenge-skill balance: The task is slightly beyond your current comfort zone but within your capability. Not too easy (boredom), not too hard (anxiety). The sweet spot.
  2. Clear goals: You know what you're trying to achieve, moment to moment.
  3. Immediate feedback: You receive real-time information on whether you're making progress.

Disrupt any of these three conditions, and flow either doesn't happen or collapses mid-session.

This is where AI enters the picture.

The Flow Cycle: Four Phases That Matter

Understanding how AI affects flow requires understanding that flow isn't a switch you flip. It follows a four-phase cycle:

Phase 1: Struggle. The loading phase. You're absorbing information, wrestling with complexity, mapping the problem space. This phase is uncomfortable. Cortisol and norepinephrine are elevated. Your brain is working hard. Most people interpret this discomfort as a signal that something is wrong and immediately switch back to shallow work (aka procrastination).

It isn't. The struggle phase is neurological priming. You're building the pattern recognition and neural scaffolding that make flow possible.

Phase 2: Release. You step back. Stop pushing. Take a walk, make a coffee, shift to a lighter task, stare outside for 30 seconds. This is when the unconscious processing that the struggle phase initiated begins to consolidate.

Phase 3: Flow. If phases 1 and 2 have done their work, the brain is primed. Challenge meets skill. Attention narrows. The work starts moving. This is the state. Game time...

Phase 4: Recovery. Flow depletes neurochemical reserves. After a deep session, you need genuine restoration before the next one. As covered in the Sleep Advantage edition and the Recovery edition, this isn't optional.

Here's the critical insight for the AI conversation: AI affects each phase differently. Knowing which phase you're in determines whether reaching for the AI tool is a smart move or a costly one.

Where AI Degrades Flow

1. It Disrupts the Challenge-Skill Balance

This is the central threat. And the one most people never see coming.

Csikszentmihalyi's (1990) research is unambiguous: flow requires challenge and skill to be both high and roughly matched. Drop the challenge below your skill level and you get boredom. Push challenge above skill and you get anxiety. The flow channel sits at that dynamic balance point.

When AI handles the cognitively demanding portion of your work (drafting, analysis, problem structuring, strategic reasoning) it removes the challenge while your skill level stays the same. The task becomes trivially easy. You're no longer in the flow channel. You're reviewing AI output, not thinking.

The Harvard Business School field experiment by Dell'Acqua et al. (2023) is the most instructive study in this space. They studied 758 BCG consultants using GPT-4 across a variety of tasks. The headline finding gets quoted constantly: 25% speed increase, 40% quality improvement. What almost nobody mentions is the other finding: for tasks outside the AI's capability frontier, consultants who used AI performed 19 percentage points worse than those who didn't.

Why? Because they had outsourced their cognitive engagement to the AI. When the AI was wrong or inadequate, they had lost the active reasoning needed to catch it. Model hallucination is not going anywhere soon.

The challenge-skill balance didn't just shift. For complex, genuinely novel problems, it collapsed.

2. It Eliminates the Productive Struggle Phase

This is the mechanism I see most clearly in the leaders I work with, and the most counterintuitive one to explain.

AI's primary value proposition, "don't struggle with the blank page, let AI start it for you", is also its greatest risk for flow. Because the struggle phase isn't uncomfortable overhead you should want to skip. It's the neurological loading that makes flow possible.

When you wrestle with a complex problem before the answer arrives, you're doing three things simultaneously: building a detailed mental map of the problem space, priming the dopaminergic reward circuits for the satisfaction of resolution, and developing the skills that will allow you to enter flow on that class of problem more reliably in the future.

The learning science literature calls this the generation effect (Slamecka & Graf, 1978): actively generating information, even when effortful and error-prone, produces dramatically better understanding and retention than passively receiving the same information. Letting AI generate the answer bypasses this entirely.

The implication is significant. Leaders who routinely use AI to skip the struggle phase may be increasing short-term output while systematically degrading the skill base and neural priming that enables flow long-term. The biological ceiling quietly lowers. Not in ways you'll notice this week. In ways you'll notice in two years.

3. It Fragments Attention Through Micro-Interruptions

As I covered in the Context Switching edition, the cost of interruptions on deep work is enormous. Mark, Gudith, and Klocke (2008) found that after an interruption, it takes an average of 23 minutes and 15 seconds to fully return to the original task. Monsell (2003) demonstrated that even anticipated, voluntary task switches carry measurable cognitive costs.

AI writing assistants that offer real-time inline suggestions, autocomplete tools that appear as you type, and AI copilots that proactively surface "relevant" information function as persistent micro-interruption generators.

Each AI suggestion requires a micro-decision: accept, reject, modify, or ignore. This micro-decision engages exactly the prefrontal cortex processing that needs to quiet for flow to occur (Dietrich, 2004). You're never fully in your own thinking. There's always a secondary stream of AI-generated content competing for your attention.

This becomes doubly true for agents. You setup an agentic process and kick it off, then you wait... Well you might as well start another one in parallel, or switch tasks until it is complete. Deep focus? Forget about it.

For the geeks: Ward et al. (2017) demonstrated that the mere presence of a smartphone on your desk, turned face-down and silent, measurably reduces available cognitive capacity. The implication for always-on AI assistants is significant. The background awareness that "I could just ask the AI" may be quietly draining working memory even when you're not actively using it.

4. It Creates Automation Complacency

The human factors research on automation complacency (Parasuraman & Manzey, 2010) is some of the most robust in cognitive science. Decades of research in aviation, nuclear power, and healthcare show a consistent pattern: when automation handles tasks reliably, humans disengage from active cognitive monitoring. They shift from doing to supervising. Refer to the paragraph above regarding agentic work.

This matters for flow because complacency is the neurological opposite of absorption. Flow requires active, fully engaged participation. Supervision of AI output is:

  • Low in challenge (reviewing is less demanding than generating)
  • Low in autonomy (you're reacting to AI output, not creating)
  • Structurally hostile to the deep absorption that flow requires

Goddard, Roudsari, and Wyatt (2012) documented this in clinical decision-making: clinicians who relied on AI diagnostic support were more likely to follow incorrect recommendations because they had disengaged from active clinical reasoning. The AI had effectively switched off the cognitive engagement that made them good at their jobs. Read that part again. The same mechanism applies at a boardroom strategy table or in a leadership team session.

5. It Can Quietly Erode Intrinsic Motivation

Self-Determination Theory (Deci & Ryan, 2000) identifies three fundamental psychological needs that drive intrinsic motivation: autonomy (feeling self-directed), competence (feeling capable and effective), and relatedness (feeling connected to something meaningful). Intrinsic motivation is both a precondition for flow and one of its key outputs.

AI can quietly undermine all three. Autonomy suffers when work feels reactive rather than generative (as your would expect, automation reduces autonomy). Competence suffers when AI routinely produces output that matches your unaided work. "Why bother, the AI does it better" is not a thought that leads to flow. Authorship suffers when the work no longer feels fully yours.

Doshi and Hauser (2024, Science Advances) found that while AI enhanced individual creativity scores, it reduced the collective diversity of novel content. AI-assisted stories were rated more creative by rubric, but were more similar to each other. The individual risk: AI-assisted work can start to feel generic. And generic work doesn't produce flow. It produces output. (When you think about how an LLM is trained and the mechanics that sit behind it's output this makes complete sense).

Where AI Supports Flow

Now for the other side. Because used intentionally, AI is one of the most powerful flow-support tools available to you right now.

1. Eliminating the Shallow Work That Blocks Flow Access

Here's the practical reality: most senior leaders aren't failing to enter flow because AI has made their work too easy. They're failing because their calendar is buried in email triage, meeting prep, status updates, routine correspondence, and administrative friction. None of which is flow-worthy and all of which consumes the time and cognitive bandwidth that flow requires (not to mention the decision fatigue).

Brynjolfsson, Li, and Raymond (2023) studied 5,179 customer service agents using an AI assistant and found average productivity gains of 14%, with the largest benefits coming from the reduction of routine task burden. When AI handles shallow work, the kind with low challenge and low meaning, it protects the cognitive bandwidth and calendar space that flow demands.

This is what leverage over volume actually looks like in practice. Not working harder or longer. Strategically removing the friction that blocks access to the work that actually matters.

The critical distinction, this one is important, is whether AI is offloading shallow work or the challenging core of your work. The former creates flow opportunities. The latter destroys them.

2. Accelerating the Loading Phase Without Eliminating Struggle

There's a distinction most people miss: there's a difference between AI eliminating the struggle phase and AI accelerating the pre-struggle loading.

A significant portion of what happens before productive struggle is information gathering: reading background material, reviewing prior decisions, aggregating data, identifying constraints. This is loading work, but it's not the struggle itself.

AI that helps you build contextual map faster (summarising background documents, identifying relevant precedents, structuring the problem space) doesn't eliminate the struggle phase. It compresses the time to reach it, so you can spend more of your cognitive prime time on the actual challenge.

Sweller's (1988) Cognitive Load Theory provides the framework here: reducing extraneous cognitive load (information retrieval, formatting, organisation) while preserving germane cognitive load (the effortful processing that produces understanding) optimises conditions for deep engagement. AI used this way is a precision tool for load management, not challenge avoidance. There's an important difference.

3. Tightening Feedback Loops

One of flow's three core preconditions is immediate feedback. This means real-time information (or as close to it as possible) on whether you're making progress. In most knowledge work, feedback is painfully slow. You write a strategy document and won't know for weeks whether it landed. You make a significant decision and the consequences unfold over months.

AI can provide proxy feedback that tightens this loop significantly: immediate quality assessment of written arguments, instant adversarial testing of strategic reasoning, real-time analysis of scenarios and models. Peng et al. (2023) found that software developers using GitHub Copilot completed tasks 55.8% faster partly because the feedback loop of working code was dramatically tightened. The equivalent in strategic knowledge work is substantial.

When AI functions as a thinking partner rather than an answer machine, it can help maintain the feedback signal that the flow state requires.

4. Protecting the Flow Environment

As covered in the Context Switching edition, the modern leader's environment is systematically hostile to deep focus. González and Mark (2004) found that knowledge workers switch activities every three minutes on average. Any tool that extends your average uninterrupted work block moves you closer to the 10–15 minute minimum threshold needed for flow onset.

AI-powered scheduling tools can actively defend deep work blocks from meeting creep. AI communication filters can distinguish genuinely urgent messages from those that can wait, reducing the vigilant inbox monitoring that fragments attention. Mark, Voida, and Cardello (2012) found that cutting off email access for five days reduced stress and increased focus significantly. AI that intelligently filters rather than eliminates communication achieves a similar effect without the cost of disconnection.

5. Using AI to Raise the Challenge

This is the most under-utilised AI application for flow, and the one I find most interesting.

Most people use AI to make their work easier. The optimal use for flow is to use AI to make your work harder in precisely the right way: specifically, to push the challenge back up toward your skill level when work starts to feel too straightforward.

Ask AI to steelman every assumption in your strategy document. Have it generate the strongest possible counterarguments to your position. Use it to identify edge cases and failure modes in your plan. Simulate an adversarial board member interrogating your decision.

This is AI as a challenge calibration tool. It uses the technology to maintain the challenge-skill balance in your favour, rather than letting it collapse into AI-assisted ease. Flow remains accessible. Your skills stay sharp. The work remains yours. Dial it up or down in real-time.

The Inverted-U: Finding Your Optimal AI Dose

Consistent with a principle you'll recognise from the cortisol edition, AI's relationship with flow follows an inverted-U curve:

Too little AI: You're buried in shallow work. Email, formatting, routine communication, information aggregation. All consuming the time and cognitive bandwidth that flow requires. Flow opportunities are crowded out before you get near them.

Optimal AI: AI handles friction, accelerates the loading phase, protects deep work windows, and tightens feedback loops. You retain the challenging, skilled, creative core of your work. Flow frequency increases.

Too much AI: AI takes over the challenging work itself. Challenge drops below your skill level. The struggle phase disappears. Skills gradually atrophy. Autonomy and competence, the foundations of intrinsic motivation, erode. Flow frequency declines. Work starts to feel efficient but hollow.

The optimal point on this curve is personal and dynamic. It depends on your current skill level, the nature of your specific work, and your relationship with challenge. But the diagnostic question is simple:

After a session with AI assistance, does your work feel more engaging or less?

If less: you've likely crossed the line.

Protocols: Using AI Intentionally for Flow

Protocol 1: The Shallow Work Boundary

The Science

AI applied to shallow work creates flow opportunities. AI applied to the challenging core of your work destroys them. This protocol draws that boundary deliberately, rather than letting it form by default.

The Protocol

  1. List the 5-7 activities that consume most of your weekly calendar. Be specific.
  2. For each, ask: "Does this require the distinctive expertise, judgement, or relationship context that makes me valuable in this role?"
  3. Activities where the answer is clearly no (email triage, meeting notes, routine correspondence, data formatting, research aggregation) are AI-appropriate.
  4. Activities where the answer is yes (strategic analysis, key relationship communication, novel problem-solving, high-stakes decisions) stay with you, without AI assistance.
  5. Apply AI systematically to the first category for two weeks. Protect the second from AI involvement entirely.
  6. Track daily flow frequency using a simple 1-5 self-rating. Compare across the fortnight.

Expected Outcome

Within 1-2 weeks, most leaders notice a meaningful increase in available deep work time and a reduction in the administrative drag that makes flow feel inaccessible. The quality of the protected work also tends to improve, because you're bringing full cognitive engagement rather than AI-assisted production.

Protocol 2: The Loading Accelerator

The Science

AI is maximally valuable in the pre-struggle phase (gathering information, building context, structuring the problem space). It is maximally damaging when used to skip the struggle itself. This protocol creates a deliberate separation between the two.

The Protocol

  1. Before any significant deep work session, identify what information or context you need to load.
  2. Use AI to compress the loading phase: summarise relevant background, identify key constraints, aggregate data, map prior decisions. Budget 15–20 minutes maximum.
  3. After the loading phase, close the AI tool. Start the actual work without AI assistance.
  4. When you hit genuine uncertainty or creative blocks, do not immediately reach for AI. Stay with the discomfort for at least 20–30 minutes. The friction is neurological priming, not a signal to outsource.
  5. Only re-engage AI after sustained effort. And when you do, use it as a thinking partner ("What am I missing?" "What's wrong with this reasoning?") rather than an answer source.

Expected Outcome

This protocol preserves the flow cycle's struggle phase while reducing the low-value pre-work that can consume your peak cognitive hours. Most leaders find that deep work sessions structured this way are noticeably more absorbed and productive than AI-assisted sessions. Expect the work to feel harder and, as a result, better.

Protocol 3: The Flow Guard

The Science

Environmental protection is one of the most underrated flow enablers. As covered in the Context Switching edition, a single interruption costs up to 23 minutes of recovery time. An AI writing assistant surfacing suggestions every few minutes is functionally equivalent to working in an open-plan office with someone regularly tapping you on the shoulder.

The Protocol

  1. Designate 2–3 deep work blocks weekly (90–120 minutes each) as AI-free periods. No inline assistants, no suggestion tools, no AI tabs open in the background.
  2. Use AI before these blocks: clarify the goal for the session, load the relevant context, anticipate potential friction points.
  3. Use AI after these blocks: clean up notes, process outputs, handle any administrative follow-through.
  4. During the block itself: single application, notifications off, AI closed. Full cognitive presence.
  5. Use an AI-powered scheduling tool (Reclaim.ai or Clockwise are solid options) to defend these blocks from meeting creep and protect them in your calendar architecture.

Expected Outcome

Most leaders who implement structured AI-free deep work blocks report a significant increase in both the frequency and depth of flow within 2–3 weeks. The contrast effect is often striking. The quality of thinking in protected blocks versus AI-assisted sessions becomes immediately noticeable.

Protocol 4: The Challenge Calibrator

The Science

Flow requires challenge to sit slightly above your current comfort zone. When AI makes work easier, it can push challenge below your skill threshold and produce disengagement even when work is technically getting done. This protocol uses AI to maintain challenge rather than reduce it.

The Protocol

After completing a significant piece of work such as a strategy document, analysis, key decision, or plan, run it through these AI prompts before finalising:

  1. "What are the three strongest arguments against this position?"
  2. "What assumptions am I making that could be wrong? Identify the most critical."
  3. "Where are the edge cases and failure modes in this plan?"
  4. "What would a highly sceptical board member ask about this?"
  5. "What important perspectives am I likely missing?"
  6. "What limiting beliefs might be at play here?"

The goal is not to have AI improve your work. It's to use AI to generate the challenges and constraints that require you to do better work. AI surfaces the questions. You do the thinking.

Expected Outcome

This protocol typically produces a qualitative shift in how the work feels to more engaging, more rigorous, and more genuinely yours. It becomes a challenge to have Claude come back and say "you've covered all the bases in report, identifying all opposing arguments, well done". It also consistently produces stronger outcomes, as the adversarial questioning surfaces gaps and assumptions that solo work tends to miss. Expect higher satisfaction in the process, and more confidence in the output.

TL;DR: The Quick Version

For my time-poor readers:

  • Flow is your highest-value cognitive state. Up to 5x more productive than normal, and the state your best strategic thinking comes from.
  • AI's default mode degrades flow by reducing challenge, eliminating productive struggle, and fragmenting attention through micro-interruptions.
  • The challenge-skill balance is the critical variable. When AI makes work too easy, you leave the flow channel. The Dell'Acqua et al. (2023) BCG study shows AI users performed 19% worse on tasks the AI couldn't handle well because they'd disengaged from active reasoning.
  • The flow cycle matters. AI is appropriate in the loading (pre-struggle) and recovery phases. It is costly during the struggle and flow phases themselves.
  • Optimal AI dose follows an inverted-U. Too little AI buries you in shallow work. Too much strips out the challenge that makes flow possible. Find your line.
  • The diagnostic question: After AI-assisted work, does the session feel more engaging or less? If less, you've crossed the line.

Your Operating System Isn't Broken. But It Needs Updating.

The leaders who will perform best in an AI-augmented world aren't those who use AI the most. They're those who understand their own cognitive architecture well enough to use AI precisely. Applying it where it removes friction, protecting the challenging cognitive work that keeps their skills sharp and their flow states accessible.

This is leverage over volume in practice. Not doing more with AI. Using AI strategically to protect your biological ceiling for the work that demands your full capacity.

Track it like you'd track any performance variable: rate your flow frequency daily (1-5 scale), note whether you used AI during that work session, and watch the correlation over 2-3 weeks. The data will tell you where your line is.

Your brain isn't broken. But the operating system needs a deliberate update for the age of AI.

If you've made it this far. Well done, this was my longest article yet (hence why it was a little late coming out) but a concept I have been wrestling with for a while. Let me know how you enjoyed it!

PS: If you're working on rebuilding your performance infrastructure, including mapping your personal flow triggers, designing your calendar around your peak cognitive windows, and creating the environmental conditions for more reliable deep work. This is exactly what I work through with clients in Second Summit Ascent. Fourteen weeks. A complete operating system rebuild. If the way you're currently working isn't giving you the access to deep work and flow that your role demands, book a free discovery call here and we'll take a look at what's actually going on.

References

Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at Work. NBER Working Paper No. 31161. https://doi.org/10.3386/w31161

Csikszentmihalyi, M. (1990). Flow: The Psychology of Optimal Experience. Harper & Row.

Deci, E. L., & Ryan, R. M. (2000). The "what" and "why" of goal pursuits: Human needs and the self-determination of behaviour. Psychological Inquiry, 11(4), 227–268. https://doi.org/10.1207/S15327965PLI1104_01

Dell'Acqua, F., McFowland III, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Harvard Business School Working Paper No. 24-013. https://doi.org/10.2139/ssrn.4573321

Dietrich, A. (2004). Neurocognitive mechanisms underlying the experience of flow. Consciousness and Cognition, 13(4), 746–761. https://doi.org/10.1016/j.concog.2004.07.002

Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290

Goddard, K., Roudsari, A., & Wyatt, J. C. (2012). Automation bias: A systematic review of frequency, effect mediators, and mitigators. Journal of the American Medical Informatics Association, 19(1), 121–127. https://doi.org/10.1136/amiajnl-2011-000089

González, V. M., & Mark, G. (2004). "Constant, constant, multi-tasking craziness": Managing multiple working spheres. In Proceedings of CHI '04 (pp. 113–120). ACM. https://doi.org/10.1145/985692.985707

Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. In Proceedings of CHI '08 (pp. 107–110). ACM. https://doi.org/10.1145/1357054.1357072

Mark, G., Voida, S., & Cardello, A. (2012). "A pace not dictated by electrons": An empirical study of work without email. In Proceedings of CHI '12 (pp. 555–564). ACM. https://doi.org/10.1145/2207676.2207754

Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140. https://doi.org/10.1016/S1364-6613(03)00028-7

Nakamura, J., & Csikszentmihalyi, M. (2002). The concept of flow. In C. R. Snyder & S. J. Lopez (Eds.), Handbook of Positive Psychology (pp. 89–105). Oxford University Press.

Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410. https://doi.org/10.1177/0018720810376055

Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv preprint arXiv:2302.06590.

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002

Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604. https://doi.org/10.1037/0278-7393.4.6.592

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4

Ulrich, M., Keller, J., Hoenig, K., Waller, C., & Grön, G. (2014). Neural correlates of experimentally induced flow experiences. NeuroImage, 86, 194–202. https://doi.org/10.1016/j.neuroimage.2013.08.019

van der Linden, D., Tops, M., & Bakker, A. B. (2021). The neuroscience of the flow state: Involvement of the locus coeruleus norepinephrine system. Frontiers in Psychology, 12, 645498. https://doi.org/10.3389/fpsyg.2021.645498

Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). Brain drain: The mere presence of one's own smartphone reduces available cognitive capacity. Journal of the Association for Consumer Research, 2(2), 140–154. https://doi.org/10.1086/691462