Learning Library

← Back to Library

AI-Engineered Focus: Redesign Your Workday

Key Points

  • AI can be used not just to increase output but to reshape work‑day conditions, reducing interruptions, speeding recovery, and aligning tasks with available time.
  • Engineer John Duruk frames productivity with three key “dials”: interruption frequency (λ), recovery time after an interruption (δ), and the length of an uninterrupted block needed for deep work (θ).
  • By measuring these three parameters you can predict whether a day will be productive, revealing that many knowledge workers are interrupted every 2 minutes and need about 10 minutes to refocus, which makes genuine deep work nearly impossible.
  • This model reframes focus as an engineering problem rather than a willpower or discipline issue—if you start the day with zero viable focus blocks, you shouldn’t expect miracles.
  • Small adjustments to any of the three knobs can produce nonlinear gains, meaning modest changes to interrupt frequency, recovery time, or block length can dramatically increase deep‑work capacity.

Full Transcript

# AI-Engineered Focus: Redesign Your Workday **Source:** [https://www.youtube.com/watch?v=UrJdtQgXnCw](https://www.youtube.com/watch?v=UrJdtQgXnCw) **Duration:** 00:15:10 ## Summary - AI can be used not just to increase output but to reshape work‑day conditions, reducing interruptions, speeding recovery, and aligning tasks with available time. - Engineer John Duruk frames productivity with three key “dials”: interruption frequency (λ), recovery time after an interruption (δ), and the length of an uninterrupted block needed for deep work (θ). - By measuring these three parameters you can predict whether a day will be productive, revealing that many knowledge workers are interrupted every 2 minutes and need about 10 minutes to refocus, which makes genuine deep work nearly impossible. - This model reframes focus as an engineering problem rather than a willpower or discipline issue—if you start the day with zero viable focus blocks, you shouldn’t expect miracles. - Small adjustments to any of the three knobs can produce nonlinear gains, meaning modest changes to interrupt frequency, recovery time, or block length can dramatically increase deep‑work capacity. ## Sections - [00:00:00](https://www.youtube.com/watch?v=UrJdtQgXnCw&t=0s) **Reprogramming Your Workday with AI** - The speaker outlines how AI can be used to control three core productivity factors—interruption frequency, recovery time, and length of uninterrupted work blocks—drawing on John Duruk’s systems‑engineering framework. - [00:03:10](https://www.youtube.com/watch?v=UrJdtQgXnCw&t=190s) **Engineering Deep Work with AI** - The speaker frames deep work as a design problem—not a moral one—explaining how modest tweaks to interruption patterns and task structuring, amplified by AI tools, can convert scattered minutes into sustained, high‑impact work blocks without increasing total hours. - [00:06:16](https://www.youtube.com/watch?v=UrJdtQgXnCw&t=376s) **AI as Focus Management Tool** - The speaker reframes AI from a generic productivity aid to a system that filters interruptions, preserves work context, and structures tasks into manageable chunks, outlining strategies for smarter AI‑driven workflow. - [00:10:56](https://www.youtube.com/watch?v=UrJdtQgXnCw&t=656s) **AI‑Assisted Task Chunking Strategies** - The speaker explains that while AI can automate microtasks such as code generation and outlining, over‑reliance can fragment one’s mental model, so employing a simple chunking method and treating deep‑work periods as a measurable service level helps maintain focus and coherence. - [00:15:01](https://www.youtube.com/watch?v=UrJdtQgXnCw&t=901s) **Building a Deep‑Work Toolkit** - The speaker suggests gathering resources into a toolkit designed to support focused, interruption‑free work, especially by minimizing Slack distractions. ## Full Transcript
0:00You know, for the first time, you can 0:01use AI not just to do more work, but to 0:04quietly reprogram the conditions of your 0:06workday, which means fewer 0:08interruptions, faster recovery, and 0:10tasks that actually fit into the time 0:12you have. The inspiration for this post 0:14is an engineering systems thinking blog 0:17from John Duruk. It's fantastic. I'm 0:20going to link to it, but I'm also going 0:21to give you like the cliff notes, 0:23two-minute summary here at the top of 0:24the video so you know what I'm referring 0:26to as I get into how I use AI to get 0:29more work. Okay. Who is John Duruk? He 0:31is a longtime engineer. He's a founder, 0:33co-founder at Felt, Exuber, Dig, etc. 0:36And he really is a systems engineer. 0:38Don't think of him as a productivity 0:39coach. And you can figure out why he's a 0:42system engineer and not a productivity 0:44coach when you see how he writes about 0:46productivity because it's all 0:47mathematically inclined. I'm going to 0:50give it to you without the need for 0:51math. Basically, he says there are three 0:53key dials that you can turn that 0:55determine your day. The first is how 0:56often you're interrupted per hour. He 0:58gives that a Greek letter he calls 0:59lambda. The second is how long it takes 1:01your brain to get back on task after an 1:03interruption. He calls that delta. and 1:06the length of an uninterrupted block of 1:09time that is enough for it to be real 1:12work. How many minutes is it for it to 1:13be real work and you're not interrupted 1:15during that time? He calls that theta. 1:17So, forget the Greek letters. You don't 1:19need to know those. But if you use those 1:21three parameters, you can actually see 1:24ahead of time whether your day is likely 1:26to be productive or not. You can look at 1:29your day and say, "Wow, I have no 1:31uninterrupted blocks." like I we have 1:33engineered the productivity out of my 1:35day and that is actually his larger 1:37point is that when you look at studies 1:39by Microsoft of all of all companies you 1:42see that for people who are heavy 1:43coordinators at work they get 1:45interrupted on the average of every two 1:47minutes and if you think about it if you 1:49get an interruption every two minutes 1:51during the workday if it takes you 10 1:52minutes to get back on task you're in 1:54negative territory every single day 1:56which explains a lot of how we all feel 1:58so the first contribution he makes is 2:00just to turn the idea of I can't focus 2:02into a model that we can talk about, 2:04engineer, and think about. And I think 2:06that's a great gift. And I would argue 2:08that this framing is actually pretty 2:09empowering because it reminds us that 2:12focus is an engineering problem. Focus 2:14is not a willpower problem. Focus is not 2:17I'm not disciplined enough, right? Focus 2:19is what is the expected number of focus 2:22blocks that are sufficient for 2:23productivity in the day. If it's zero at 2:25the start of the day, I shouldn't expect 2:27magic. The other thing I want to call 2:29off is that this is a model that is 2:32susceptible to nonlinear benefits from 2:35small changes, which is a fancy way of 2:37saying if you care about getting more 2:40deep work done, you should tweak the 2:42knobs of your day pretty aggressively 2:44because even small changes can lead to 2:46really significant upside for you. So, 2:48as an example, the same 155 minutes of 2:51focus can yield four units of work if 2:54your theta length, if your deep work 2:56length is 30 minutes, but only three 2:58units if your theta length is 45 3:00minutes. But if you tweak that length of 3:03focus just a little bit, you can squeeze 3:05in another unit of work at 45 minutes. 3:07It's not that far away. Tiny shifts in 3:10lambda and delta. Tiny shifts in the 3:12number of interruptions and how long it 3:14takes to come back on task can flip days 3:16from statistically no deep work to three 3:19real blocks of work without really 3:21increasing your hours work. And really 3:23that comes back to the idea that deep 3:24work is a design choice. It's not a 3:26moral high ground, right? It's not a 3:28moral bar. If your internal standard for 3:30real work is I need 90 uninterrupted 3:33minutes and your job statistically only 3:35gives you 20 or 30 minute chunks, your 3:38capacity is mathematically forced to 3:40zero almost every day. You can respond 3:42by lowering your standards. But there's 3:44a much more interesting move here. You 3:46can keep your theta honest for the hard 3:49work and redesign tasks so more of your 3:52contribution can be done in smaller and 3:54well scaffolded pieces. And that's where 3:56tools and AI start to matter. And so 3:58this is not an ex a sort of video about 4:01a magical workplace that none of us work 4:03in where interruptions cease. I don't 4:05want you to take that away. This is 4:07actually a very practical video where we 4:09look at work and focus as an engineering 4:11problem and then ask ourselves how AI 4:13can be a super lever that helps us to 4:16move that entire work system into a more 4:18positive environment for ourselves. The 4:20fourth thing that he calls out in the 4:22blog post that I think is really 4:23relevant for us to keep in mind as we 4:24get into the AI portion of this, he 4:26makes it obvious that the the 4:29interruption level is a culture setting, 4:32not a personal trait. Right? So Lambda 4:34is driven by meeting norms. It's driven 4:36by DM etiquette. It's driven by Slack 4:38channel sprawl. Right? It's driven by 4:40just a quick question behavior. If you 4:42are a manager, that can feel like really 4:44positive news because the biggest 4:46productivity lever in the model is 4:48something you can change via your social 4:50norms. You don't have to beg people to 4:52be more disciplined. You can just choose 4:54not to slack them. You can choose to 4:56leave them be. All of this sets the 4:59stage for AI in a way that feels useful. 5:03AI becomes interesting because it can 5:05help us turn the dials at scale. It 5:07doesn't just give us one more tab to 5:09work on. Now you might wonder why does 5:11AI belong in this picture at all? 5:12Because more AI at work stories usually 5:14jump to look the model can do stuff. I 5:17would argue that John Duruk's model 5:18invites a very different question that's 5:20more useful. If the limiting factor on 5:22our deep work is these three variables 5:25of lambda, delta, and theta. How often 5:28we're interrupted, when we come back to 5:29it, how long it takes, and and how long 5:31our deep work takes. Where can AI 5:33actually usefully push those numbers in 5:35the right direction? Interestingly 5:37enough, AI is often unusually good at 5:40exactly the things that sit around those 5:42three knobs. It's good at monitoring and 5:45routing. So, it can watch streams of 5:46messages. It can classify urgency. It 5:48can decide what gets through. That's 5:50something we've actually seen in 5:51startups that are starting to declutter 5:53the inbox on exactly those principles. 5:55It can summarize and recall, right? It 5:57can compress past context into something 6:00that you can reload very quickly and 6:01efficiently so you don't miss something, 6:03but it doesn't interrupt you. It can 6:04also decompose and scaffold out very 6:06easily, right? It can turn big fuzzy 6:08tasks into smaller executable ones, 6:10which is one of the things that Dudo 6:12calls out as a big hack around theta. 6:14And so instead of AI as a productivity 6:16boost abstractly, I want you to think of 6:19AI as a focus system tool. I want you to 6:22think of AI as a tool that helps you 6:24choose when and how often people are 6:26allowed to knock on your door. or AI as 6:29a tool that remembers the work state you 6:31had so your brain can reboot quickly and 6:33doesn't have to do a full reload. If 6:35you've ever loaded up a past chat, GPT 6:38chat and scanned it and said, "Now I 6:39know where I am." You've done this. For 6:41theta, this is about changing the shape 6:44of the work so it can fit into more 6:46finite blocks of time. It's like carving 6:48it into useful chunks. This is a much 6:50more useful way of thinking about AI and 6:53productivity than adding a chatbot to 6:54Slack, guys. So let me give you a few 6:56strategies that come out of this for me. 6:59And if you're wondering, yes, I actually 7:01use these strategies. What I am giving 7:03you is both both the theoretical 7:05framework that Duro outlines and also my 7:08personal productivity approach that I 7:10have derived based on optimizing my own 7:13productivity settings with AI. So 7:15strategy number one, use AI for fewer, 7:17smarter interruptions. The obvious play 7:19is just to stop notification firewalls. 7:22An agent can sit on Slack, Teams, and 7:24email and auto answer trivial questions. 7:26It can bundle non-urgent pings. It can 7:28break through in real time only when it 7:30really matters. I do this all the time. 7:32It doesn't even have to be a super uh 7:34aggressive AI as superhuman has an AI 7:36that looks at what's important and 7:38what's not important. That helps me a 7:40lot. That's not super hard to set up. 7:41You just set up your superhuman 7:43instance, right? Same for meetings. An 7:45AI scheduler agent can autopose async 7:47updates. It can route status checks to a 7:49doc. and it can push back on your 7:51calendar spam by default. Again, this is 7:54often built into good email clients. 7:56It's increasingly something that you can 7:58get out of box. Now, there are real 8:00trade-offs here. You are making a 8:02conscious trade to have slower replies 8:04in the occasional mclassified email or 8:07the occasional mclassified Slack ping in 8:09exchange for fewer total interruptions. 8:11You are taking some risk. Some people 8:13will read a slower response as somewhat 8:16standoffish, but at the end of the day, 8:18if you're getting deep work done, the 8:20trade-off you're making is that the 8:21actual productivity will be worth it. I 8:23realize that's not true for everyone, 8:25but for many of us, being able to do the 8:27deep work is what leads to the 8:28transformational benefit both for our 8:30own mental wellness and also, frankly, 8:32for the things that we're working on, 8:33the company we're working for, or even 8:35if we're working for ourselves. So, 8:37strategy one is really use AI any way 8:41you can to shut off interruptions. And 8:42there really are a lot of tools. I I've 8:44mentioned superhuman, but lindy.ai helps 8:46you with this. There are other tools out 8:48there as well. Uh, and I'm going to 8:50assemble a whole list for the Substack 8:51that will help you on the the 8:53productivity and interruption side so 8:55that you can actually focus. Strategy 8:57number two, use AI to shrink your delta 9:00to get back into the problem faster. Use 9:03it to load context more quickly. At the 9:06simple end, you can just ask the model, 9:08what was I working on last? And because 9:10most models now remember past 9:11conversations, that works well. Claude 9:13does. Chat GPT does. At the more agentic 9:16end, you could actually set up a context 9:18agent that snapshots what you're editing 9:20and reading and comes back with a task 9:22log. I haven't personally felt the need 9:24to go that far. I find that if I can 9:26search through my past chats and I have 9:28kept good notes and I can reload that 9:30context quickly, it is good enough. It 9:33depends on you and what you need to boot 9:36your brain back with context. The key is 9:38making sure that you consciously 9:40remember to ask for the context you need 9:43to boot quickly and that you constantly 9:46note. Whether it's through vocal sort of 9:48granola notetaking or whether it's 9:50through typing or whether it's through 9:53summarizing in your handwriting in the 9:55notebook, right? Whatever it is, make 9:56sure you get something that reduces your 9:59future reload time. And I wasn't kidding 10:01about the notebook. I have a physical 10:02notebook and if I need to remember what 10:04I was doing, I can flip the page very 10:06quickly to two days ago. And as funny as 10:08it sounds, that's not necessarily AI, 10:10but it does reboot that context very 10:12quickly. And of course, if I want to, AI 10:14can also take a picture of that, read 10:17it, and give me a summary of what was 10:19useful from the day before. One of the 10:20ways I've actually used that is when 10:21I've had a page of handwritten notes in 10:23a meeting, and I'm like looking through 10:25it, and I can't find what I'm looking 10:26for because my handwriting is so bad. AI 10:29handwriting recognition is good enough 10:30now that I can take a picture and I can 10:33get the AI to read my handwritten notes 10:36for me and say, "Oh, that was the thing 10:37you were thinking about." Helps reload 10:39context fast. Strategy three, use AI to 10:43fit more work into realistic real world 10:46blocks. So, if your minimum time to do 10:48deep work is 90 minutes and you have 10:49very few of those blocks, can you chunk 10:51your work into 20 to 40 minute chunks? 10:53AI is really helpful here. The model can 10:56generate tests and logging and 10:57boilerplate when it comes to code. We've 10:59talked about that. The model can do 11:01outlines for writing. The model can 11:03structure headings. It can do research 11:05for you. The model can do a first pass 11:07on a document. Now, if you take this too 11:10far, you can end up with a day that's 11:12composed of so many AI assisted 11:14microtasks that you have no mental model 11:16of the whole problem anymore. And then 11:18you lose your human taste, right? You 11:20can also have AI that doesn't decompose 11:22correctly. So the chunks are not the 11:25right size for actual deep work. But 11:27what I have found in practice is that 11:29the chunking strategy is actually one of 11:31the easiest to employ here. Like you 11:33might have to install a tool to not get 11:35interrupted. You might have to really 11:38think about how to get back into the 11:40problem quicker and do deep different 11:41note-taking strategies. But for using AI 11:44to make work chunkable, that's as simple 11:46as saying, I have this whole thing to 11:48do. Give me some ideas to chunk it. 11:50Right? Like it's actually a very 11:51effective way forward. Duruk's most 11:53powerful idea is that leaders should 11:55treat focus like uptime for engineers. 11:57So they should define service levels for 11:59deep work blocks and manage toward them. 12:01I think that's really powerful if you're 12:03managing an engineering team. I also 12:05think it's completely unworkable if you 12:07are in some other job roles because 12:09other job roles your job is the meeting, 12:11right? But I think the idea of taking 12:14focus seriously and measuring against it 12:17is still meaningful. And one of the 12:19things AI can help us with here is 12:21learning to read a calendar and actually 12:24measure our deep work. And so we are at 12:26a point now where if you color code your 12:28calendar and you tell an AI, please read 12:31this calendar for my deep work blocks, 12:33it can do it. You can also extend that 12:35very easily into a vibecoded app for 12:37your whole team. Or you can do it 12:39without a vibecoded app just by grabbing 12:41screenshots and loading them in with a 12:43good prompt. I'm going to build a prompt 12:45for this. It is not difficult to 12:47actually measure and I think Duruk has a 12:49point that what we measure we care about 12:51and we can start to think about if we 12:54work on teams how we optimize for deeper 12:56work. If you put this all together, you 12:58get a pretty simple menu, right? At the 13:00individual level, we can use prompts, we 13:03can use built tools, we can use simple 13:05automation to get into context quicker 13:09and to lower our effective theta, right? 13:11To make it easier by decomposing 13:13problems with AI to get more work done. 13:15We can also slightly reduce our 13:17interruptions by using personal 13:19notification rules or maybe some simple 13:22ways of working that reduce interruption 13:25continually over time. It's as simple 13:26sometimes as turning off Slack, right? 13:28It's not all AI. But at the team level, 13:31we also need culture changes. And that's 13:33something that we can start to advocate 13:35for. Especially if you're in management, 13:36this is something you can just start to 13:38roll out to help your team. You can 13:40agree on Slack and meeting norms that 13:41aim to target less interruptions. You 13:44can adopt shared resumption patterns 13:47such as every spec, every PR has a 13:51here's where to pick this up section and 13:53AI helps to maintain it. That's helpful 13:55for engineers. You can have similar 13:56rituals on the non-technical side where 13:59you have here's how to ramp into this 14:01context at the top of a particular page 14:03if someone has to pick up work. The 14:04thing that I want to leave you with, 14:06this has been key for my productivity 14:08and it's something that came out a lot 14:09in this in this essay by Duruk as well 14:12and it's something that I think AI 14:13really helps us with. You do not have to 14:16treat your focus as a mystical personal 14:19trait that some people have and AI is 14:21the shiny add-on over the top. I hear 14:24this a lot from people. People will say, 14:25'Nate, how do you get so much done? And 14:27they treat me like I'm a magical person 14:28with magical AI. I'm not. Treat your 14:31attention like a system with dials and 14:34treat AI as a lever that helps you to 14:36turn those knobs more efficiently first 14:38for yourself and then for your team. And 14:40if you work in lead an org, eventually 14:42for your whole org. Empowerment is not 14:44really about I try harder in this 14:45situation. It's about I understand the 14:48system that leads to focus and deep work 14:51and I have a set of AI enabled levers I 14:54can start pulling. That has been my goal 14:56with this video. I'm going to put some 14:57prompts together to help you with that. 14:58Get some tools together. I want this to 15:01be a toolkit that enables deep work for 15:04you. Best of luck with uh actually 15:06getting work done and not getting 15:07interrupted by Slack.