--- title: "Hypertasking: Productive Multitasking in the Age of AI" description: "Hypertasking is the art of running AI agents while clearing your task list. Learn how to restructure your working day to get 3-5x more done with AI." canonicalURL: https://aaronlynn.com/ai/hypertasking/ date: 2026-07-07T00:00:00.000Z --- Any time you have a conversation with work colleagues or business friends now, there is some discussion of AI. This recently happened to me and the range of responses was staggering. One of them said, "Oh man, I'm falling behind — I can't keep up with all the new tools and workflows being released." Another said, "Yeah I tried out Grok on my phone, it was pretty great." And a third shrugged and said they'd heard about it in the news. This is the world of AI right now. AI has come from some people's working lives hard and fast. For others... it's an idle curiosity — something they hear about between the weather and the sports news. ![Dot-grid infographic where each dot is \~3.2 million people: of 8.1 billion humans, about 84% have never used AI, 16% use a free chatbot, 0.3% pay for AI, and only 0.04% use AI coding tools][image-1] For me, the way I work has changed beyond recognition in the last 12 months.[^1] Claude now touches everything I do. Writing, coding, ops, research, project management, content creation, marketing... all of it. I still remember the first time I hit the five-hour usage limit on my Claude 5X Max plan. This was during some heavy coding/vibe coding. I messaged a friend — "Now what?" He said to work on something else. So I opened up my marketing work. Oh wait, I use Claude for that too. Opened up project management. Could do it manually, but it would take five times as long. Pretty much everything I switched to doing, relied on Claude. My friend's advice in the end: "Go hang out on the balcony and enjoy the sunlight." And the blog post you are reading right now? My entire [writing process][1] has changed. It used to be: start with an idea, contribute notes over days, OCD-check my content over weeks until it felt right. Now I have a hyper-intelligent thinking partner who can bring a dozen perspectives to bear on my notes before I even start writing. It can also write in a way that sounds like me, proofread better than I can, generate images, format and upload everything online.[^2] My take is this: **If you are not using AI and maximising your usage of it, you are missing out and falling behind.** Note: If you are behind, or if you have no idea what Claude is, grab my [AI for Work Productivity][2] course and start there. I will not be covering the basics of how to use Claude, Cowork, Projects or Claude Code in this article — that is what the course is for. --- ## The Golden Age of AI We are living in the Golden Age of AI right now. And most people are not taking advantage of it. I believe that AI currently has the value of a $250,000/year employee for $100-200/month. That is not hyperbole. That is what I experience when I use Claude Code, Codex and Gemini across my professional and personal work.[^3] It will not be this inexpensive forever. People complaining about usage limits on the $20/month plan do not realise how cheap it is — they are trying to squeeze blood from a stone. As [Matt Shumer puts it][3], it is like using a flip phone today and saying that mobile phones suck. You want to maximise your AI running time right now, while it is _practically free._ ### But Wait — AI Seems Pretty Fast to Me You might be reading this and thinking: _"What are you on about, Aaron? AI gives me answers in seconds!"_ If you are using Claude on the web, sure. Responses come back quickly, and you are not going to be sitting around long enough for any of this to matter. Treat that as a [deep focus task][4] and move on. But if you are working with AI beyond the web — in Claude Code, Claude Cowork, Codex or other CLI tools — it is a different story. I'm talking about: - Using parallel subagents to write 20 newsletters at the same time: **5-minute wait**. - Create a website mockup in html/css with the exact copy that will be used: **10-minute wait**. - Code a new microservice: **15-minute wait**. - Run a multi-agent adversarial code review: **20-minute wait**. - Autonomous overnight app creation: **5-hour wait**.[^4] These are the people this article is for. Non-tech knowledge workers using Claude Cowork or Claude Code. Developers and vibe coders running multiple sessions. Anyone whose AI work involves periods of waiting. If you are only using AI on the web or on your phone, you probably do not need hypertasking. Yet. ## The Problem: AI Thinking Is Slow OK, not really. AI is still faster than you and I. But it is not instant. You prompt something, and there is a 2-5 minute gap for non-coding tasks. Sometimes 10-30 minutes for coding tasks. And during that gap... you sit there. ![16-bit Street Fighter-style bonus stage: a fighter smashes a smartphone and the YouTube, Instagram, Reddit and Wordle icons to rubble beneath "destroy distractions, unlock focus" and "stay focused" signs, while an "AI thinking… 30%" progress bar loads below][image-2] What do most people do in that time? - Just wait. - Check their phone. - Scroll and browse endlessly on Instagram, TikTok, Reddit... - Watch YouTube. - Brain teasers, touch typing practice, and... Wordle.[^5] - Get a coffee. All of that is wasted time, except the coffee. It is [destructive snacking][5] dressed up as a break. We need something better. That something is called **Hypertasking**. Hypertasking is the art of running multiple, concurrent AI agents at full speed while you clear your task list in the gaps between prompts. To get there, let us recap how we currently do deep work and why traditional multitasking is bad. ## Recap: What About Deep Work and Focus? I have a full article on [how to deep work and single task][4], and this is how productive people have worked for years. Break up your tasks into periods of deep focus, start at number one and work your way through. It is absolutely the best way to get things done fast and at high quality. Good knowledge workers get in four hours of deep work a day. Productivity masters can stretch that to eight. But this is now outdated. Because we now live in the age of AI, and a new paradigm is required. So do we give deep work up and just... multitask? No. ## Recap: So Is Multitasking Back in Vogue? Multitasking is still a bad idea. You are not "multitasking" anything, the human brain does not work that way. What you are doing is **rapid switch-tasking**: Rapidly going from one context to another. The research is pretty clear on this: **Task-switching costs you up to 40% of your productive time.** A 2001 study found that executive control processes in task switching cause significant performance loss, even when the tasks are familiar.[^6] In a real work environment, **it takes 23 minutes to refocus after an interruption.** Interrupted work leads to higher stress and faster but lower-quality output.[^7] And for those people who like to watch YouTube or TikTok while they work: **Media multitasking physically reduces grey matter, the part of your brain responsible for cognitive control.**[^8] Overstimulation. Grey matter reduction. Impaired memory encoding. Dopamine addiction. Your cortisol goes up. Your glucose stores deplete faster. If it is not clear: multitasking = bad. ### And what about The Molecule Man? I wrote about a version of potentially productive multitasking in my [Next Level Productivity][6] article called **The Molecule Man**. A way to actually do **switch-tasking** with minimal downsides. But you have to be GOOD at productivity to pull it off. All of these apply: - [God-tier productivity skills][7]. - Knowing how to mentally switch contexts and note down where you left off. - A clean workbench (clean up and reset your work environment at the end of each major task or day). - Artefacts and systems in place to track what is going on. [Task lists][8], project management tools, [journals][9], awareness of your energy and [hero mode][10] for the day. - Technological mastery. You are going to be doing a lot of window switching. - In the article I said that multiple monitors help. With hypertasking, multiple monitors are **mandatory**. Here is the thing: AI is just a [force multiplier][11]. If you are not already productive and organised, it is not going to make you more of either. It will just amplify whatever bad habits you already have. The Molecule Man as I described it was v1.0. With AI, we have **Molecule Man v2.0: Hypertasking**. A form of rapid delegation to AI, where [your AI agents act as fast, hyper-intelligent staff][12]. ## How to Hypertask The aim of hypertasking is to maximise the runtime of **hyper-intelligence**. That is my generic term for the current-generation of AI models.[^9] You want to run these close to 24/7. For coders, this is called [hyperengineering][13]. Same idea, different name. You want to use this hyper-intelligence as much as possible while it is practically free. This is more than just _tactics._ This requires rewriting your entire working day's _strategy._ ### The three-segment journal system ![16-bit arcade "Mode Select" screen with three daily modes — Deep Focus (a figure meditating in flames), AI-Assisted (directing researcher, writer and analyst robots), and Low-Level tasks cleared in the gaps][image-3] I use a [journal system][9] built around three outcomes per day. If you just have tasks rather than outcomes, group your tasks into three segments: **1. Deep Focus Tasks** Block off one to two hours a day for these, first thing in the morning or whenever you have the most cognitive resources. No AI unless it is part of the task. These are tasks that require thinking, creativity, logic, intelligence. You want to zero in, focus, get them done and get out with no interruptions. This is [classic deep work][4]. **2. AI-Assisted Work** If you are a developer or vibe coder: - Start the generation running. - Use multiple git worktrees. I find about three is my cognitive limit as a newish developer.[^10] If you are a non-tech knowledge worker: - Whatever work is AI-heavy. Done in Claude Code, Codex or Claude Cowork usually. - You will probably max out at two concurrent tasks because responses come back faster when code is not involved. Aim to have AI-assisted work running all hours outside of your deep focus block. **3. Low-level Switchable Tasks** These are the tasks you can quickly switch to and work on while the AI is running. Easy, 5-10 minute tasks you can fragment and that do not require deep focus. You should be able to clear all of these by the end of the day while your AI work runs. Examples: - Admin tasks. - Clearing email. - Answering Slack messages and reminders. - Prioritising project backlogs. - Small content pieces. - Manual uploads of anything. - Returning phone calls. - Home stuff and errands.[^11] ### A sample day ![16-bit arcade versus screen: "Hypertasker" player one, backed by AI coder, writer and researcher panels marked complete, faces off against "The Backlog" player two — a fanged monster made of an overflowing task list with a 999+ inbox and urgent deadline alerts][image-4] Here is what a hypertasking day looks like: | Time | Activity | | ------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 8:00am - 10:00am | Deep work. No AI, no interruptions. | | 10:15am - 12:00pm | Start AI delegation — set up prompts and agents, let them run. Clear admin tasks, email, phone calls, project management while AI works. Check in on AI outputs and approve work periodically. | | 12:00pm - 1:00pm | Lunch break. | | 1:00pm - 30 minutes before end of day | Continue AI delegation cycles + remaining low-level tasks. Review AI outputs, give feedback, re-prompt as needed. | | End of day | Final reviews, wrap up, have AI create handoff documents, plan for tomorrow. | A friend of mine runs a book publishing business. Before AI, his workflow looked like this: spend days delegating research, crafting detailed outlines, handing them off to a human writer, waiting a couple of weeks for a draft, reviewing it, then sending it back with notes, then waiting again. Rinse and repeat. The bottleneck was always the human writer's bandwidth. I helped him set up AI agents for his publishing pipeline. Now his workflow is: review an outline the AI drafted (15 minutes), approve it or give feedback, then go do something else entirely — answer emails/messages, return calls, review another outline, or clear low-level admin work. If he needs some downtime, he can pop out for lunch or just grab a coffee. The AI writes the full draft while he is off doing other things. When it is done, he reviews the output, tells the AI to edit what is needed, and moves on to the next one.[^12] That is hypertasking in action. He is not sitting there watching the AI write. He is clearing his entire task list in the gaps between checking outlines and reviewing outputs. The AI does most of the heavy lifting — he just steers. ### No more 8-hour deep work days You are purposely interrupting the flow of one task to clear other tasks. Why? To maximise AI as a [force multiplier][11] and get more done in less time. If you run this properly, you WILL get 3-5 times more done than you used to: - **Writing email newsletters?** Used to be a week-long or multi-week process. Now compressed into a couple of 15-minute sessions. - **Writing a sales letter?** One week down to one day. - **Heavy data analysis?** One week with a team of analysts down to one hour with you and Claude. - **Coding an app?** Used to take a month. Now takes a week or two because the syntax is handled for you. 3-5x more output. Same hours. That is both incredible and insane if you think about it. ### Isn't this just regular multitasking? _"But Aaron, didn't you just say that multitasking is bad?"_ I did. And it is. But hypertasking is structurally different: 1. **You are switching at self-initiated break points.** You are not being interrupted by external forces. AI will not complain if you don't answer it right away. 2. **AI is doing the heavy cognitive lifting.** You are not trying to hold two complex thought processes in your head simultaneously. 3. **You are delegating, not being interrupted.** Just as with a human staff member, you choose when to respond to your AI agents. 4. **You are switching to low-cognitive-effort tasks.** You are not trying to juggle four deep focus items at the same time. It turns out humans can do what researchers call **threaded cognition** — the management of concurrent tasks when those tasks draw on different cognitive resources.[^13] Delegation uses one cognitive resource. Low-level admin tasks use a different one. AI handles all of the execution — writing, coding, researching. You handle direction-setting and review. That said, you WILL experience some of the downsides of multitasking. More on that in a moment. ![16-bit versus screen comparing multitasking (a panicking worker, system-overload errors, focus 10%, stress 95%, grey matter draining) with hypertasking (a calm worker backed by AI agents, focus 85%, output 300%, stress 40%)][image-5] ### Isn't this just what developers have always done while code compiles? Tech people have done "work while compiling" for decades. But AI delegation is _cognitive work,_ not just mechanical compilation. It is more like delegating coding to a junior developer and waiting for them to report back to you. You can get on with other things while you wait. But the output you get back requires review, feedback and sometimes re-prompting, not just a pass/fail check. ## How to Deal with the Cognitive and Physical Demands of Hypertasking There is one major downside to hypertasking: **The cognitive and physical toll on your body is significant.** You are being hyper-productive. Literally putting out 3-5 times what you did before. AI handles the grunt work of implementation, but you still do the mental heavy lifting — reviewing all of it, switching between contexts, making decisions, giving feedback, re-prompting. And there is a dopamine hit because you actually are doing productive things.[^14] [This is a common experience amongst people who start working this way.][15] So how do you handle this? ### 1. Good old-fashioned productivity advice You handle it by using some good, old-fashioned personal productivity advice. This is where the [time management stack][7] and [productivity stack][16] earn their keep. **Use [timeslices][17].** Work for 37 minutes, take an 8-minute break. Get up. Drink water. Walk around. Stretch. Do not be like the people I see at coworking meetups glued to their seats for five hours straight. **Get on top of your health and energy.** This means [rituals][18], exercise, and diet. Four cups of coffee does not count as a nutrition plan. **Lean on your systems.** Having established [goals][19] and [a journal system][9] keeps you grounded when everything is moving fast. **[Plan your days the night before][20].** And then remain fast and agile as the day unfolds. **Mandatory downtime.** One hour a day at night MINIMUM. Half a day to a full day off per week. You WILL burn yourself out if you skip this. This is non-negotiable. **Learn to [delegate to humans too][21].** Not everything needs to be done by an AI or needs to be managed by you. ### 2. AI-Specific tooling advice These layer on top of your existing productivity skills: **Understand what a project directory or repository is** so you can put everything the AI needs in one place. You do not want to waste time finding files or granting permissions between prompts. **Use Auto Mode or set up your Claude Code permission settings properly** so you are not sitting there pressing "yes" all the time. Configure your sandbox and allow/deny lists. Ask Claude to help you set this up. **Build a memory system** for AI repos you work in frequently. This means the AI knows what to do and accumulates your organisational knowledge over time, and you stop having to give the same feedback repeatedly. Start with CLAUDE.md or AGENTS.md, then build your own markdown system (ask Claude Code to build it for you) or use something like [claude-mem][22]. **Learn to use git if you handle code.** Set up AI autocommit so you have granular version history. Developers call these atomic commits. If something goes sideways, you can always roll back. OR, use Google Drive if it’s just plain documents and images (though git remains better). **Spend a couple of hours a week updating your workflow.** AI can help with this. The goal is that you manage less and AI manages itself more over time. ### 3. Worst case? Slow it down. Go back to deep focus without the low-level tasks in-between for a day or two. While you do want to maximise your usage of affordable AI hyperintelligence, you don't have to do it every single day. ![Productivity Stack][image-6] Hypertasking is a systems tool in your [productivity stack][16], it isn't a religion. If you want to take a break and do an 8-hour deep focus day, do that. ## What To Do Next This is the new way of working. As I keep telling people I meet — the way I work today is completely different from the way I did six months ago. People who adopt hypertasking will get 3-5 times more done than their peers over the next year. Promotions, raises, new business... these will go to the people who learn to work alongside AI, not the people who wait for it to come to them. And the people who don't adopt it? I expect they will fall behind. Here is what you should be doing next: 1. **Grab an AI premium plan** and start using AI beyond the free tier. I'd recommend the Claude Code 5X Max plan at $100/month, but you can start at the $20/month plans. That's the cost of a couple of flat whites. 2. **Structure your day into the three segments.** Deep focus, AI-assisted work, and low-level switchable tasks. Use your [journal][9] to plan this the night before. 3. **Start with one AI task running in the background** while you clear admin work. Scale up to two or three AI tasks as you get used to the system. 4. **Monitor your energy and attention** and take breaks. Hypertasking is demanding. Respect that. _Note: Still playing catchup with AI itself? Grab my [Complete Claude, Claude Code & AI for Work Productivity][2] course. It is built for non-tech business people who want to get started properly._ _Note: Already hypertasking but feeling the heat? Grab [Next Level Productivity][23] and make sure your productivity fundamentals are rock solid._ [^1]: And I don't mean in a subtle, incremental way. I mean I look back at how I operated pre-AI and it seems... quaint. Like watching someone use a fax machine. [^2]: After apologising to my assistant who no longer helps with any of this. [^3]: Mostly Claude Code at the time of writing. But the field moves fast and that could change. [^4]: For this one you can just go to sleep. [^5]: Wordle. While Claude writes your microservice. That is genuinely what some people are doing. Sigh. [^6]: Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). Executive control of cognitive processes in task switching. _Journal of Experimental Psychology: Human Perception and Performance_, _27_(4), 763–797. https://doi.org/10.1037/0096-1523.27.4.763 [^7]: Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. _Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '08)_, 107–110. https://doi.org/10.1145/1357054.1357072 [^8]: Loh, K. K., & Kanai, R. (2014). Higher media multi-tasking activity is associated with smaller gray-matter density in the anterior cingulate cortex. _PLOS ONE_, _9_(9), e106698. https://doi.org/10.1371/journal.pone.0106698 [^9]: Claude Fable 5, Opus 4.8 and GPT-5.5 at time of publication. [^10]: Your mileage may vary. Some people run five or six. [^11]: This is one of the many benefits of [working from home][14]. You can throw a load of laundry or take a quick break in between prompts. [^12]: Yes, AI can write a consistent, 40,000-word book in about 30 minutes. [^13]: Salvucci, D. D., & Taatgen, N. A. (2008). Threaded cognition: An integrated theory of concurrent multitasking. _Psychological Review_, _115_(1), 101–130. https://doi.org/10.1037/0033-295X.115.1.101 [^14]: Usage of AI tools — especially Claude Code — is genuinely addictive once you realise what you can do with it. It takes away the grunt work of implementation, helps with the cognitive overhead of strategy and planning, and lets you create things you could not before because of a lack of technical knowledge. Hello, gateway drug 🙃 [1]: https://aaronlynn.com/productivity/writing-system/ [2]: https://aaronlynn.com/courses/ai-for-work-productivity/ [3]: https://shumer.dev/something-big-is-happening [4]: https://aaronlynn.com/productivity/how-to-deep-work-and-single-task/ [5]: https://aaronlynn.com/productivity/eliminating-destructive-snacking-and-distractions/ [6]: https://aaronlynn.com/productivity/next-level-productivity/ [7]: https://aaronlynn.com/productivity/time-management-stack/ [8]: https://aaronlynn.com/productivity/personal-task-management/ [9]: https://aaronlynn.com/productivity/journals/ [10]: https://aaronlynn.com/productivity/energy-management/ [11]: https://aaronlynn.com/productivity/force-multipliers/ [12]: https://aaronlynn.com/courses/ai-agents-guide/ [13]: https://hyperengineering.bottlenecklabs.com/p/no-xp-waste [14]: https://aaronlynn.com/business/remote-working/ [15]: https://www.reddit.com/r/ClaudeCode/comments/1r6y7od/mental_burnout_from_too_many_parallel_claude_code/ [16]: https://aaronlynn.com/productivity/productivity-stack/ [17]: https://aaronlynn.com/productivity/timeslices-flowtime-pomodoro-technique/ [18]: https://aaronlynn.com/productivity/morning-rituals-evening-rituals/ [19]: https://aaronlynn.com/productivity/effective-goal-setting/ [20]: https://aaronlynn.com/productivity/wtf-days/ [21]: https://aaronlynn.com/business/effective-delegation-situational-leadership/ [22]: https://github.com/thedotmack/claude-mem [23]: https://aaronlynn.com/courses/next-level-productivity/ [image-1]: https://assets.aaronlynn.com/data/blog/ai/hypertasking/ai-adoption-dot-grid-infographic.webp [image-2]: https://assets.aaronlynn.com/data/blog/ai/hypertasking/hypertasking-destroy-distractions-bonus-stage.webp [image-3]: https://assets.aaronlynn.com/data/blog/ai/hypertasking/hypertasking-three-mode-select-screen.webp [image-4]: https://assets.aaronlynn.com/data/blog/ai/hypertasking/hypertasker-vs-backlog-versus-screen.webp [image-5]: https://assets.aaronlynn.com/data/blog/ai/hypertasking/multitasking-vs-hypertasking-comparison.webp [image-6]: https://assets.aaronlynn.com/data/blog/productivity/productivity-stack/the-productivity-stack.webp > Related course: AI for Work Productivity — $147 — /courses/ai-for-work-productivity/