Nate B Jones applies an engineering model of focus to explain how AI can reduce interruptions and recover lost deep work time
A solo presentation by Nate B Jones on using AI as a lever to improve deep work conditions, inspired by an engineering systems blog.
Summary
Nate B Jones presents a framework drawn from a systems engineering blog post by John Duruk, which models workplace productivity using three mathematical variables: interruption frequency, recovery time after interruption, and the minimum block length required for meaningful work. Jones argues that focus is an engineering problem, not a willpower problem, and that small changes to these variables can produce disproportionately large gains in productive output. He then maps specific AI tools and strategies onto each of the three variables, showing how AI can reduce interruptions, speed up context recovery, and decompose large tasks into chunks that fit realistic time windows. The video closes with a call to treat attention as a designed system rather than a personal trait.
Key Takeaways
FULL TRANSCRIPT
Introduction: Why this video exists
Nate B Jones: For the first time, you can use AI not just to do more work, but to quietly reprogram the conditions of your workday — which means fewer interruptions, faster recovery, and tasks that actually fit into the time you have.
The inspiration for this video is an engineering systems thinking blog post by John Duruk. It's fantastic. I'm going to link to it, but I'm also going to give you the cliff notes — a two-minute summary here at the top so you know what I'm referring to as I get into how I use AI to get more work done.
John Duruk's model: the three variables of productive work
Who is John Duruk? He is a longtime engineer and founder — co-founder at Felt, Uber, Digg, and others. He really is a systems engineer. Don't think of him as a productivity coach. You can figure out why he's a systems engineer and not a productivity coach when you see how he writes about productivity, because it's all mathematically inclined.
I'm going to give it to you without the need for math. Basically, he says there are three key dials you can turn that determine your day. The first is how often you're interrupted per hour — he gives that a Greek letter he calls lambda. The second is how long it takes your brain to get back on task after an interruption — he calls that delta. And the third is the length of an uninterrupted block of time that is enough for it to be real work: how many minutes does it need to be for it to count as real work, during which you're not interrupted? He calls that theta.
Forget the Greek letters — you don't need to know those. But if you use those three parameters, you can actually see ahead of time whether your day is likely to be productive or not. You can look at your day and say, "Wow, I have no uninterrupted blocks" — like, we have engineered the productivity out of my day. And that is actually his larger point.
When you look at studies by Microsoft, of all companies, you see that for people who are heavy coordinators at work, they get interrupted on average every two minutes. And if you think about it — if you get an interruption every two minutes during the workday, and if it takes you ten minutes to get back on task, you're in negative territory every single day. Which explains a lot about how we all feel.
Focus as an engineering problem, not a willpower problem
So the first contribution Duruk makes is just to turn the idea of "I can't focus" into a model that we can talk about, engineer, and think about. And I think that's a great gift. I would argue that this framing is actually pretty empowering, because it reminds us that focus is an engineering problem. Focus is not a willpower problem. Focus is not "I'm not disciplined enough." Focus is: what is the expected number of focus blocks that are sufficient for productivity in the day? If it's zero at the start of the day, I shouldn't expect magic.
The other thing I want to call out is that this is a model that is susceptible to nonlinear benefits from small changes — which is a fancy way of saying that if you care about getting more deep work done, you should tweak the knobs of your day pretty aggressively, because even small changes can lead to really significant upside.
As an example: the same 155 minutes of focus can yield four units of work if your theta — your deep work block length — is 30 minutes, but only three units if your theta is 45 minutes. But if you tweak that block length just a little, you can squeeze in another unit of work. It's not that far away.
Tiny shifts in lambda and delta — tiny shifts in the number of interruptions and how long it takes to come back on task — can flip days from statistically no deep work to three real blocks of work, without really increasing your hours worked. And that comes back to the idea that deep work is a design choice. It's not a moral high ground. It's not a moral bar.
If your internal standard for real work is "I need 90 uninterrupted minutes," and your job statistically only gives you 20 or 30-minute chunks, your capacity is mathematically forced to zero almost every day. You can respond by lowering your standards. But there's a much more interesting move: you can keep your theta honest for the hard work, and redesign tasks so more of your contribution can be done in smaller, well-scaffolded pieces. And that's where tools and AI start to matter.
Where AI fits into the focus model
So this is not a video about a magical workplace where interruptions cease. I don't want you to take that away. This is a practical video where we look at work and focus as an engineering problem, and then ask ourselves how AI can be a super lever that helps us move that entire work system into a more positive environment.
The fourth thing Duruk calls out in the blog post — and this is really relevant as we get into the AI portion — is that the interruption level is a culture setting, not a personal trait. Lambda is driven by meeting norms. It's driven by DM etiquette. It's driven by Slack channel sprawl. It's driven by "just a quick question" behavior. If you are a manager, that can feel like really positive news, because the biggest productivity lever in the model is something you can change via your social norms. You don't have to beg people to be more disciplined. You can just choose not to Slack them. You can choose to leave them be.
All of this sets the stage for AI in a way that feels useful. AI becomes interesting because it can help us turn the dials at scale. It doesn't just give us one more tab to work on.
Now, you might wonder why AI belongs in this picture at all, because most "AI at work" stories usually jump to "look, the model can do stuff." I would argue that Duruk's model invites a very different question that's more useful. If the limiting factor on our deep work is these three variables — how often we're interrupted, how long it takes to come back, and how long our deep work blocks need to be — where can AI actually usefully push those numbers in the right direction?
Interestingly, AI is often unusually good at exactly the things that sit around those three knobs. It's good at monitoring and routing — it can watch streams of messages, classify urgency, and decide what gets through. That's something we've actually seen in startups that are starting to declutter the inbox on exactly those principles. It can summarize and recall — it can compress past context into something you can reload very quickly and efficiently, so you don't miss something but it doesn't interrupt you. It can also decompose and scaffold very easily — it can turn big, fuzzy tasks into smaller executable ones, which is one of the things Duruk calls out as a big hack around theta.
So instead of thinking of AI as a productivity boost abstractly, I want you to think of AI as a focus system tool. Think of AI as a tool that helps you choose when and how often people are allowed to knock on your door. Or AI as a tool that remembers the work state you had, so your brain can reboot quickly and doesn't have to do a full reload. If you've ever loaded up a past ChatGPT chat and scanned it and said, "Now I know where I am" — you've done this.
For theta, this is about changing the shape of the work so it can fit into more finite blocks of time. It's like carving it into useful chunks.
This is a much more useful way of thinking about AI and productivity than just adding a chatbot to Slack.
Strategy one: use AI for fewer, smarter interruptions
So let me give you a few strategies that come out of this for me. And yes, I actually use these strategies. What I'm giving you is both the theoretical framework that Duruk outlines and also my personal productivity approach, derived from optimizing my own productivity settings with AI.
Strategy number one: use AI for fewer, smarter interruptions. The obvious play is to build notification firewalls. An agent can sit on Slack, Teams, and email and auto-answer trivial questions. It can bundle non-urgent pings. It can break through in real time only when it really matters. I do this all the time. It doesn't even have to be a super aggressive AI setup — Superhuman has an AI that looks at what's important and what's not. That helps me a lot. That's not super hard to set up. You just configure your Superhuman instance.
Same for meetings. An AI scheduler agent can propose async updates, route status checks to a doc, and push back on calendar spam by default. Again, this is often built into good email clients and is increasingly something you can get out of the box.
Now, there are real trade-offs here. You are making a conscious trade to have slower replies and the occasional misclassified email or misclassified Slack ping, in exchange for fewer total interruptions. You are taking some risk. Some people will read a slower response as standoffish. But at the end of the day, if you're getting deep work done, the trade-off is that the actual productivity will be worth it. I realize that's not true for everyone, but for many of us, being able to do the deep work is what leads to the transformational benefit — both for our own mental wellness and, frankly, for the things we're working on, the company we're working for, or even if we're working for ourselves.
So strategy one is really: use AI any way you can to shut off interruptions. I've mentioned Superhuman, but Lindy.ai helps with this too. There are other tools out there as well.
Strategy two: use AI to shrink your delta and reload context faster
Strategy number two: use AI to shrink your delta — to get back into the problem faster. Use it to load context more quickly. At the simple end, you can just ask the model, "What was I working on last?" And because most models now remember past conversations, that works well. Claude does. ChatGPT does.
At the more agentic end, you could set up a context agent that snapshots what you're editing and reading and comes back with a task log. I haven't personally felt the need to go that far. I find that if I can search through my past chats and I've kept good notes and can reload that context quickly, it's good enough. It depends on what you need to boot your brain back with context.
The key is making sure you consciously remember to ask for the context you need to boot quickly, and that you constantly take notes — whether it's through voice note-taking with something like Granola, or typing, or summarizing in your handwriting in a notebook. Whatever it is, make sure you get something that reduces your future reload time.
And I wasn't kidding about the notebook. I have a physical notebook, and if I need to remember what I was doing, I can flip the page very quickly to two days ago. As funny as it sounds, that's not necessarily AI, but it does reboot context very quickly. And of course, if I want to, AI can also take a picture of that, read it, and give me a summary of what was useful from the day before.
One of the ways I've actually used that is when I've had a page of handwritten notes from a meeting and I'm looking through it and can't find what I'm looking for because my handwriting is so bad. AI handwriting recognition is good enough now that I can take a picture and get the AI to read my handwritten notes and say, "Oh, that was the thing you were thinking about." It helps reload context fast.
Strategy three: use AI to fit more work into realistic blocks
Strategy three: use AI to fit more work into realistic real-world blocks. So if your minimum time to do deep work is 90 minutes and you have very few of those blocks, can you chunk your work into 20 to 40-minute pieces? AI is really helpful here. The model can generate tests, logging, and boilerplate when it comes to code. The model can do outlines for writing, structure headings, do research for you, and do a first pass on a document.
Now, if you take this too far, you can end up with a day composed of so many AI-assisted microtasks that you have no mental model of the whole problem anymore — and then you lose your human taste. You can also have AI that doesn't decompose correctly, so the chunks are not the right size for actual deep work.
But what I've found in practice is that the chunking strategy is actually one of the easiest to employ. You might have to install a tool to avoid getting interrupted. You might have to really think about how to get back into the problem quicker and try different note-taking strategies. But for using AI to make work chunkable, it's as simple as saying, "I have this whole thing to do — give me some ideas to chunk it." It's actually a very effective way forward.
Measuring and managing focus like uptime
Duruk's most powerful idea is that leaders should treat focus like uptime for engineers — define service levels for deep work blocks and manage toward them. I think that's really powerful if you're managing an engineering team. I also think it's completely unworkable in some other job roles, because in those roles your job is the meeting. But the idea of taking focus seriously and measuring against it is still meaningful.
One of the things AI can help with here is learning to read a calendar and actually measure deep work. We are at a point now where if you color-code your calendar and tell an AI "please read this calendar for my deep work blocks," it can do it. You can also extend that very easily into a vibe-coded app for your whole team, or you can do it without a vibe-coded app just by grabbing screenshots and loading them in with a good prompt.
I'm going to build a prompt for this. It is not difficult to actually measure, and I think Duruk has a point that what we measure we care about. We can start to think about, if we work on teams, how we optimize for deeper work.
Putting it all together: individual and team levels
If you put this all together, you get a pretty simple menu. At the individual level, we can use prompts, built tools, and simple automation to get into context quicker and to lower our effective theta — to make it easier by decomposing problems with AI to get more work done. We can also slightly reduce our interruptions by using personal notification rules or simple ways of working that reduce interruption continually over time. It's as simple sometimes as turning off Slack. It's not all AI.
But at the team level, we also need culture changes. That's something we can start to advocate for — especially if you're in management, this is something you can just start to roll out to help your team. You can agree on Slack and meeting norms that aim to target fewer interruptions. You can adopt shared resumption patterns — such as every spec and every PR having a "here's where to pick this up" section, with AI helping to maintain it. That's helpful for engineers. You can have similar rituals on the non-technical side, where you have a "here's how to ramp into this context" section at the top of a page if someone has to pick up work.
The core mindset shift
The thing I want to leave you with — this has been key for my own productivity, and it came out a lot in Duruk's essay as well, and it's something AI really helps with. You do not have to treat your focus as a mystical personal trait that some people have, with AI as the shiny add-on on top. I hear this a lot. People will say, "Nate, how do you get so much done?" and they treat me like I'm a magical person with magical AI. I'm not.
Treat your attention like a system with dials, and treat AI as a lever that helps you turn those knobs more efficiently — first for yourself, then for your team, and if you lead an organization, eventually for your whole org.
Empowerment here is not about "I try harder." It's about "I understand the system that leads to focus and deep work, and I have a set of AI-enabled levers I can start pulling." That has been my goal with this video.