I set up an automated AI news feed, then switched it off. The material arrived, but I did not read it: I had no attention to spare, and another digest interested me less than my own work. The automation reliably created one more place I was supposed to check. I had already done enough to receive more information; I still needed to understand why I wanted to know it.
Yet my work requires me to understand what is changing. I conduct audits, help companies change their engineering processes, adopt AI and train people. A new model or way of organising agents can affect what I should recommend to a client. Sometimes an announcement brings this to my attention; sometimes I visit a company and see how people have already configured a tool I know almost nothing about. That gives me a specific question, and I begin researching it.
In episode six of 3 AImigo, I talked with Alexander Polomodov and Evgeny Sergeev—Sasha and Zhenya—about living with this constant flow while continuing to do useful work. At the time of our conversation, Sasha was in the second week of a sabbatical, a long break from work, after many years of continuous work, including applying AI in software development (AI4SDLC) at T-Bank. Zhenya was an Engineering Director at Flo Health, working on AI adoption in development and advising startups. Each of us had our own interests, obligations and ways of allocating attention.
3 AImigo, episode six. Watch the recording on YouTube
Zhenya and Sasha turned out to organise incoming information quite carefully, while I limit it more and often start from a practical need. But even a good filter leaves another question: when will we think about what we have received? An agent can prepare research, a day’s lessons or a report. The person still has to choose a direction, understand the material and decide when they have done enough for today.
Our conversation gradually moved beyond news and tools. Changes make us reconsider our working roles, and new possibilities can easily leave us feeling that we use too little of what is available. Sasha’s choice led him to change the way he lives. Zhenya is still looking for ways to bring more of the hands-on work he enjoys back into his job, and frankly acknowledges that some of the tension remains. For me, working out new things myself helps, when I leave time for it.
First, Choose Where to Look
Zhenya began with a question that is easy to skip once a new tool has appeared: what matters to me right now? We still have twenty-four hours in a day, work and a life alongside it. Even recording a podcast during Sasha’s sabbatical is a chosen priority to which he has given time. Naming our interests first lets us filter out some of the incoming material before we start devising automation.
For Zhenya, that choice connects to his regular planning: he thinks about his day in the morning, and considers the week separately. At the time of our conversation, his longer planning horizon had narrowed to one or two months, which was enough for him for now. He had read Maxim Dorofeev’s Jedi Techniques years earlier and draws on Getting Things Done (GTD), a system for organising tasks and regularly reviewing them. The official GTD overview separates capturing incoming material from clarifying its meaning, organising reminders, reviewing and taking action. In Zhenya’s adaptation, the move from “something interesting has appeared” to “what am I going to do with it?” is particularly clear.
A stream of changes appears before the compass. After it, only material relevant to the chosen direction is shown; the rest is set aside. Commitments and time for life bound the available attention.
Changes
My question
Set aside
Chosen direction
Sasha suggested first estimating how much attention is available at all. Family, children, work obligations, development and personal interests all need some of it. It helps when you can study and try something new on tasks you are paid to do. If the subject that interests you falls outside those tasks, it needs additional time that you still have to find. Because he was not working at the time, he sometimes had more than ten hours for his own pursuits. Even then, by evening he reached the point of “enough already”.
That detail matters to me. Choosing a direction does not itself make the day longer. Building a system that works means fitting an interest around your obligations and the energy left after them. Otherwise, the agent will understand what I like and bring me so much relevant material that I stop reading it again.
In our conversation, I outlined three ways of responding to change. We can keep using earlier knowledge for a long time; reconsider our approach after a crisis or another major event; or regularly ask what has changed and whether the old way still works. That last approach matters especially to someone who has chosen to stay at the forefront of AI. Constantly discovering that familiar practices need reconsidering is demanding, so recovering our energy has to become part of how we work.
Three equally weighted lines show an established approach, occasional revisions after major events and frequent reviews. All lead to the question of which cadence the chosen direction needs.
Established approach
Review after an event
Regularly question the approach
○ Meaningful change
Sasha reminded us that information overload predates AI. We used to joke that another JavaScript framework would be released before a conversation ended. With AI, we have encountered the same problem of choosing again. His reference to Attention Is All You Need, the 2017 paper on the Transformer architecture, became a play on human attention in our conversation. When we have only a few minutes for difficult material, a short explanation can help us begin studying it. We still need time afterwards to understand it.
How often we reconsider things also depends on the work we have chosen. One person needs to keep testing new possibilities; another can revisit their direction every six months or once a year. I see a risk when someone stops noticing changes altogether because their familiar position is comfortable. That does not mean everyone has to change their entire way of life every day.
Zhenya: Capture a Thought and Return to It
Zhenya likes to listen first, then read and discuss. He used to lack someone with whom he could work through a substantial research text. Zhenya now sometimes discusses the material with an agent. He explains how he understood what he read, and receives questions or observations about what he missed. The conversation helps him test his own explanation.
He organises collection around selected topics and sources. His agent setup prepares research and helps turn some of the material into audio for his morning dog walk. When an interesting thought appears in a podcast, Snipd lets him save a segment to return to later. Zhenya described a headphone press that works for him; the specific control depends on the device and its settings.
When a thought catches your interest, finding the right moment after the podcast can be difficult. The saved audio and its transcript let you listen again, make a note and add it to your personal knowledge base. Over breakfast, Zhenya reviews what has been prepared and decides what needs research or discussion, what he already understands and what he can remove because it does not relate to his chosen topics.
A path links selected topics, prepared audio, a saved fragment, a note with its source and a conversation. A new question can return Evgeny to research.
Selected topics
Research and audio
Podcast fragment
Note and source
What did I understand?
New questionReturn to the selected topics and research
This kind of system also has a cost. Zhenya tried Obsidian and became so absorbed in configuring it that he spent a great deal of time on courses, materials and a complicated setup. Eventually, he realised that his own complicated setup did not suit him. His time was going into a system meant to make working with the material easier.
He now uses ordinary plain-text files with Markdown formatting, including files for his first rough thoughts. An agent updates the material and its connections on a schedule. Zhenya referred to Andrej Karpathy’s LLM Wiki approach: keeping source material alongside an interlinked knowledge base that a person maintains with a model’s help. In Karpathy’s description, the person remains involved in selecting material and reviewing updates; the idea itself allows for different implementations.
In a note, Zhenya wants a short summary, information about its last update and freshness, and the original material on which it rests. Those fields help him decide whether to rely on the prepared explanation or return to the source. For someone who simply wants to save and discuss a useful thought, a clear page like this can be enough. Further configuration makes sense if it helps him return to the material.
The left page shows Evgeny's own complex setup, with courses and many configuration links. The right page shows a simple note with an update date, a short summary and original material. Both are assessed by what he returns to and understands.
Evgeny's complex setup
Courses and configuration
Markdown text files
- Updated
- Short summary
- Original material
Sasha: A Radar That Changes with His Choices
Sasha also uses automation to filter news, but expects it to add to his understanding. He specifically asks it to find things that expand the picture or contradict an earlier view. A digest that merely confirms his existing position does not meet that request.
Useful news differs from person to person. One is interested in ways of organising agents’ work; another cares about infrastructure and new processors. Sasha described his version of a technology radar: topics he will definitely read about, topics where he watches changes accumulate, and topics he can set aside for now. His feedback moves them between these states. It is his own adaptation of the Thoughtworks Technology Radar, whose rings also express a judgement about the attention a technology currently deserves.
Sasha is wary of a single memory that knows everything about his life. He prefers separate repositories, file stores for different activities, kept alongside one another and connected when needed. Several news items can prompt deeper research, whose results then move into an article project. The movement can go the other way too: a topic comes up in a podcast and is worth adding to the radar. Questions from his work change what the system watches next.
A radar on the left has read, watch and set-aside states. Separate research, article and podcast folders are on the right. Arrows carry news to research and an article, and a question from a podcast back to the radar. Feedback can change a topic's state.
↓ Addition or contradiction
- Read
- Watch
- Set aside
From watching to reading
Research
Article
Podcast
A new question from the podcast enters the radarFeedback
He uses similar rules for recurring checks. For example, a message arrives about a problem with analytics, search or indexing. He can give an agent read-only access, ask it to retrieve the data, investigate and prepare a report. In the morning, Sasha returns to the new event with material ready to examine.
For an agent to do some of the routine work without his constant involvement, it needs data, rules about permissible changes and checks on the result. He then periodically reviews what the agent has done. This frees attention that each step of the repeated work used to demand. Alongside it, though, sits a very different case: the agent has prepared a lesson, and Sasha must learn it. We cannot expect the same thing from a lesson as from a report.
The Tutor Prepares the Lesson; You Still Have to Learn
After deciding to move to London and choosing what he wanted to work on, Sasha put together a programme planned to run for about two and a half months. The plan allocates five hours a day. At ten each morning, the bot sends the day’s plan: theory, videos, problems, calculations and lab exercises. Sasha uses Codex and Claude; he built this automation in Codex, within his research project.
He decided to study how AI systems are built, from the bottom up: from hardware and data centres to models and an AI platform. He usually approached it in the opposite direction, so now he wants to understand its foundations. In one lesson, they worked on problems involving data-centre power and cooling, incident checks and what to look out for. The chosen goal connects the programme’s individual topics: it becomes clearer why these are the subjects to study today.
The upper loop runs from a recurring check through read-only data to a report and periodic owner review. The lower path runs from a learning goal to a plan, a question about understanding and practice. The person can finish the day while some of the plan remains.
Recurring check
Read-only access to data
Investigate and prepare a report
Periodic review
The report is prepared under rules and checks
Learning goal
Lesson plan
What do I need to understand, and why?
Practice
Return to the learning goal with a question about understanding
Even with a plan like this, the material may not add up to understanding. The tutor sends fifty pages of certification-level text; Sasha reads five and feels like closing it all. He then asks what exactly he needs to learn and why the material is in the programme. After hearing the explanation, he suggests moving to practice. The conversation lets him change how he studies while preserving the goal.
I like that the learner can challenge the plan in this example. Including a text in a schedule does not make reading it useful. If the connection to the goal has been lost, it needs restoring, or a different exercise needs choosing. A neatly organised lesson can leave the old lack of understanding intact when someone simply works conscientiously through the items.
At another point, the agent reminds him at ten in the evening that he still has an hour and a half left of the five hours and needs to add a few things. Sasha has spent the day programming, designing and reading papers. He replies: “My dear friend, that’s it for today.” He feels he has covered enough of this topic and has no energy left to continue.
Five hours sounds impressive, but this story belongs to his sabbatical. What matters more to me is that Sasha decides for himself to end the day, even though part of the plan remains. An agent can always find another task, clarification or check. A person has their own sense of enough, and a limit beyond which they end the day.
Leave Room for Your Own Thought
My way of encountering new things looks less like a study schedule. I read a few Telegram channels, including Sasha’s, look at LinkedIn and learn a great deal through companies. During an audit or implementation, I encounter a way of configuring something, a useful tool or an interesting domain. A practical task gives me a reason to investigate something I might never notice in a general digest.
I also limit my own desire to watch every Stanford and MIT course and read every interesting Anthropic paper. At some point, I noticed that ready-made explanations were beginning to take the place of my creative thinking. Sometimes I first need to sense a question that has not yet taken shape in words, then look for explanations and compare them with sources.
With GPT-6 Astra, I looked at the announcement, benchmarks and impressions on LinkedIn, where I quickly encountered sweeping judgements. Then I opened Codex and began trying the model on my own tasks. I cannot hand all the work over to long sequences of agent actions without my participation. I have a limited budget for tokens, the small pieces of text used to measure model usage, and I cannot guarantee the result I need. When I try the model myself, I can see how it works on my tasks and check what has come out of the work.
News and a question from practice lead to a question about what Aleksei's work needs. Investigating a question and trying a tool appear beside each other. Astra in Codex is an example of a trial. Work delegated to an agent remains constrained by the token budget and lack of a guaranteed result. The observation ends with a question about what Aleksei noticed.
News
A question from practice
What does my work need?
Investigate the question
Try the tool
For example: Astra in Codex
What did I notice?
Sometimes I get a feeling I call fog. There seems to be something unusual, but trying to put it into words immediately gets me nowhere. I then stop trying to express it for a while. I walk in the forest, meditate, sleep or switch to other work, such as an education programme for a client. Later, I return and find that I can begin to articulate it.
Active work comes next. I have a practice of freely writing down tasks, which I adapted from Maxim Dorofeev’s ideas. Sometimes a thought appears among the tasks, and I begin developing it in writing, talking to myself and clarifying what it relates to. In Dorofeev’s own explanation, recording tasks and things worth considering helps us hold less in our heads; my writing sometimes continues as an exploration of one of them.
It reminds me of a small stone showing through the mud. Through writing or dialogue, I gradually examine and shape it until I can see more clearly what I noticed. That brings the next question: how can I apply the thought, and what does it change in my work and life? Once it takes a clear form, I can discuss it with another person or use it in a project.
An irregular path runs from an unusual, unarticulated signal through shifting attention and returning to writing or conversation. It may lead to an articulated idea and a trial in practice; a separate dotted branch leaves the question open.
Unusual signal
No words yet
Shift attention, then return
Writing or conversation
An articulated idea
↓ Try it in my work
Still unclear
An idea may not take shape
That happened with Fable. I had the impression that something was being left unsaid in the communication, discussed it with a CTO I know, and later encountered similar reports from other users. What stayed with me was that, before reading other people’s explanations, I had noticed and formulated a hypothesis of my own. The observation does not establish what was happening inside the model; for my work, what matters is the question I managed to see for myself.
With Astra, that process was still continuing at the time of our conversation. Conflicting impressions made me think about what people at OpenAI mean by artificial general intelligence (AGI), what new capabilities change for companies, and what an engineering programme should now teach. I still had to work out which parts of the old approach remained useful and which needed reconsidering. That kind of thought gradually becomes material I can share.
Pauses are also studied separately from our own experiences. In an experiment by Baird and colleagues, an undemanding intervening activity helped participants produce more original answers to creativity tasks they had already encountered. There was no such gain for new tasks. What interests me here is that the result depended on returning to a question already begun. In my own work, I also return after a pause to what I am trying to understand and attempt to put it into words.
Draw It, Build It, Explain It
Sasha needs a pen to think. He writes a great deal by hand on a large tablet, drawing entities and arrows between them. Text unfolds sequentially, while a diagram can show the relationships he wants to understand all at once. He could not draw beautifully, but he could describe the picture in his head. An agent can now help turn that description into an image that is easier to show someone else.
Sasha also preserves time for his own thinking. News and research give him material, but continuously taking in new texts leaves little room for a thought of his own. He walks too: a birch grove, dogs, children, playing with a ball. Sometimes a question that interests him resurfaces alongside ordinary life.
Zhenya deliberately reads slowly and tries to choose books in English. Another language slows his comprehension a little. He makes handwritten and visual notes, trying to express how he understood a thought himself. It need not become immediately useful at work: sometimes preserving it matters, and he manages to return to it later.
Paths from the original material lead to a relationship sketch, a small system test and preparation to explain it to others. A question about missing understanding leads back to the material.
What I read and hear
Three ways to work with the material
See the relationships
Build and test
Explain to someone
What am I still missing?↑ What I read and hear
If you want to learn system design, watching many videos still leaves much of the practice ahead. Zhenya suggests trying to build a system, find its points of failure and understand how to check its behaviour. Those questions arise when working on a concrete system. At the same time, you have to learn adjacent skills that a prepared explanation may have left out.
When a subject he cares about rarely appears in his current work, preparing a conference or meetup talk helps him. Explaining it to people requires him to understand it himself; retelling someone else’s text does not provide the substance he wants to share. The material passes through his own experience, giving the talk its particular character. Even a solution that many people discuss can be presented through a question he encountered himself.
Zhenya has also stopped listening to podcasts while running. He now leaves those thirty or forty minutes after work to himself. Sometimes he comes back with a thought and immediately goes to write it down. In the morning, audio helps him encounter material; in the evening, the absence of new audio offers a different possibility. The same free interval need not always be used to consume information.
That brings me back to my own confidence. I have sometimes reached a thought before encountering it in other people’s publications. Behind those moments lies varied experience: engineering, management, architecture, large companies and startups. If I leave time for focus and active thinking, I have reason to trust my ability to understand. Perhaps I will reinvent something someone has already done. The understanding I reached myself will still stay with me.
Science Fiction Helps Us Ask a Different Question
During the conversation, Sasha took an unexpected detour into science fiction. Stories about agents lead him to imagine parallels that once seemed further removed from everyday work. For example, we might imagine humanity exploring the world ever more deeply until it reaches the boundary of its environment. Beyond it, humanity discovers whoever launched the simulation to produce synthetic data and long ago went off to do something else. It is an imaginative story that lets us look at the familiar progress of research from the other end.
At DeepTechNight, Sasha discussed superintelligence with a former colleague. His interlocutor assumed there had to be a single superintelligence. Sasha asked why they could not imagine a swarm pursuing a common goal, recalling fiction about such systems. The conversation gained a new question because a familiar assumption stopped looking like the only possibility.
Two thought-experiment sketches show a single large circle as one entity and a connected group with a shared goal as a possible swarm. Question marks mark them as alternatives to consider.
Familiar assumption
One entity?
Another possibility
A swarm? · Shared goal
This way of using imagination resonates with me. While we explore possibilities, how something will actually work remains an open question. Imagination helps us notice which option we have accepted by default and look for others. We can then return to the specific subject: which assumption should we test, and what would change in our work if it turned out to be too narrow?
When Personal Experiments Move Ahead of the Job
For Sasha, a major turning point came at the end of 2025. As long as AI felt like an amplifier for a person—suggesting the next piece of code or rewriting a selected block—familiar management work remained recognisable. But once the tools available to him began carrying out whole tasks without a person participating at every step, different questions arose about managers’ and engineers’ roles.
Agents had long been part of the conception of artificial intelligence. Sasha recalled Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig, which discusses agents and their environments near the beginning of the book. He also mentioned AlphaEvolve, DeepMind’s research example combining program generation with automated evaluation. For Sasha, the change was that the idea of an agent became applicable in tools he could use himself.
He began asking how this changes management and the organisation of work. If an agent can complete a substantial part of a task, what do the manager, engineer, product manager and analyst now do? In earlier technological changes, people used capital and other resources; these new agents make us reconsider how work is allocated. Soon, Sasha began asking himself what he could do and whether he was ready to change along with it.
Within Alexander's personal experience in late 2025, two cases are compared. In the first, AI helps with individual steps and the person continues the work. In the second, an agent carries out a whole task without the person taking part in every step. Alexander's question about what changes in his role remains beside it.
AI helps with individual steps
The person continues the work
An agent carries out a whole task
Without the person taking part in every step
What changes in my role?
By then, Sasha had an established life, management experience and a long-term incentive programme stretching years ahead. He could have kept his familiar way of working. Learning new things, though, had always mattered to him. During changes at the company, he handed over his earlier activities and agreed to take responsibility for AI4SDLC. He first tried to develop the topic that interested him as far as possible within his role and its existing constraints.
Gradually, what he saw in the world and did at home diverged from what he could do at work. He watched people who had started from roughly the same point move ever further ahead. His home experiments and work obligations required different ways of acting, and moving between them became psychologically difficult. Sasha described this through the image of two separate states.
Eventually, he chose the opportunity to pursue what interested him and took a sabbatical. At the time of the conversation, he felt well and was studying his chosen subjects. The fear of falling behind remained, but weighed on him less. He wanted to board one train and head towards his goal, instead of standing at the station watching every train pass.
Two labelled routes. Alexander's runs from late 2025 through AI4SDLC, a growing gap between home and work and his chosen sabbatical. Evgeny's experience with Opus during the holidays leads to open questions about practical work inside or outside the company. Energy is limited in both cases, and the fear of falling behind remains.
Alexander
Late 2025
AI4SDLC in the company
Home and work move at different speeds
Chose a sabbatical
Evgeny
Late-2025 holidays
Model and working environment
More practical work?
Inside the company or elsewhere?
Zhenya also named the holidays at the end of 2025. The version of Claude Opus available then, together with the tools configured around it, let him see a different scale of automation. He could already organise research and sequences of tasks that agents perform one after another. He began thinking about how to free time for work that interested him.
At the time of our conversation, Zhenya wanted to bring more direct participation in building systems back into his working role. Work occupies much of his life, and if the practice he finds most interesting happens only outside it, he wants to change that balance. He was considering a Member of Technical Staff role involving direct hands-on engineering, opportunities within his own company and what other companies in London offered. Questions remained at that point, and his final decision was still ahead.
What Remains of the Fear of Falling Behind
Trying to understand his strengths, Sasha saw his ability to learn about different areas that interested him. He did not consider himself the best in each one; his broad interests also made it harder to focus on one for a long time. His accumulated experience, though, helped him transfer knowledge, immerse himself deeply and reach a level at which he could contribute in his chosen field.
Once he had chosen a direction and organised his movement towards it, the fear weakened. The question of speed remained, but “study even more” no longer fitted easily into the day. That limit matters: we can be moving where we want to go and still notice people moving faster.
Zhenya once found it helpful to be advised to compare himself with who he was yesterday, rather than with the latest collection of exceptionally capable people around him. He tries to notice his own improvement and separately consider what work interests him. He particularly enjoys creating from zero to one: a new product, a first platform or a solution to an unfamiliar problem. He also knows how to expand what already exists from one to a hundred, but prolonged, steady operation tends to tire him.
Three equally weighted paths show the participants' different responses. Moving in his chosen direction helps Alexander, but the question of speed remains. Evgeny notices his own progress, while unused possibilities still trouble him. Aleksei connects the easing of his own fear with needed work and understanding its purpose. Time and energy are limited.
Alexander
Chosen direction
Moving towards the goal
Less fear; the question of speed remains
Evgeny
Compare myself with my past self
Notice my own progress
Unused possibilities still trouble me
Aleksei
Do the work I need to do
Understand its purpose
In my experience, the fear recedes
In the evenings, Zhenya keeps a diary and asks whether he was busy or moved something important forward. Days of meetings and coordination can be full to the brim yet leave little progress of his own. When too many such days accumulate, the diary gives him a reason to reconsider whether he is heading where he wants to go and what he should change at work. It connects his assessment of the day with his choice of route.
I liked that distinction because I also notice a tendency in myself to create a convincing appearance of work. Asteya is the principle of non-stealing. I apply it to myself as refusing to create that appearance. If I cannot see how an action connects to the result I need, I would rather rest, walk in the forest or enjoy a quiet meal. That brings me back to why I am continuing to act.
Yet even after making a useful choice, Zhenya still feels the weight of unused possibilities. AI makes it easy to imagine many parallel projects. Management experience has made context switching a familiar skill for him, although its demands have not disappeared. He sees some engineers who are accustomed to working in a single flow struggle with the move to several parallel activities and experience severe exhaustion.
Evgeny's diary places questions about being busy and making meaningful progress beside each other. Many days with little progress bring him back to his route. A separate line shows Aleksei asking whether an action connects to the result he needs. If he cannot see that connection, he can choose to stop and rest. Available energy and life outside work limit further demands.
Evgeny
How did the day go?
Being busy
Progress on what matters
Many days with little progressReconsider the route
Aleksei
Can I see a connection to the result I need?
Needed work
I can stop and rest
Zhenya’s main difficulty is not having enough “battery” to use all the available possibilities. He has not yet learned how to deal with that and has accepted the limit for the sake of life outside work. He understands his interests and knows how to organise his work, but the unpleasant feeling of having done far less than was possible remains.
It also helps to distinguish between kinds of switching. Sophie Leroy’s work on attention residue describes how thoughts about an unfinished task can persist after a transition and interfere with the next piece of work. For me, that is another reason to treat a quiet pause around one question differently from endless moves between several. The developer survey From Gains to Strains examines demands and resources in the context of generative AI use; its associations come from an observational study. Pressure from expectations at work therefore also deserves attention alongside personal habits.
I have noticed in myself that the fear of falling behind recedes when I am fully engaged in work I need to do. In the conversation, I even said that I had no energy left to be afraid. Alongside that experience, though, stand Sasha’s “that’s it for today”, Zhenya’s limit and my refusal to act simply to stay busy. Making progress helps me when I understand its purpose. I still do not know what proportion of the available possibilities I have used, or how much more I could have done.
Change the Route, Keep Clear Rules for Choosing
Zhenya connected our topic with product work. When ideas are plentiful, a person has to decide what to take on and what to leave aside. Sasha created his learning agent after choosing a goal. He first articulated the direction for himself, then for the tool. With that help, it is easier for him to follow his direction, even though he will miss some possibilities and someone else will get there earlier.
Sasha recalled enjoying keeping every road open as a child. He learned many things, received praise and felt capable of a great deal. At the Moscow Institute of Physics and Technology, fitting everything in for every subject became difficult; work later brought a different criterion—the result he could deliver. Breadth remained useful, but moving forward required choosing where to focus.
We can now ask an agent about almost any subject. We can take an interest in anything, which makes the question of our own direction more pressing. Sasha suggested a main route with backup plans B and C. A technical manager, for example, might consider an engineering path, particularly where skills beyond typing speed and remembering syntax would be useful. A backup route can reveal another way to apply experience already gained.
Around those options remain values, principles, external obligations and the time a person allocates to their interests. After a major change, plan B can become the main plan, while the old main plan moves into third place. Some constraints can also be reconsidered. The method of choosing remains useful: understanding what you want, comparing it with reality and using tools to find a way forward.
A compass sits within values, principles, commitments and available time. The main route and plans B and C can swap priorities.
Commitments and available time can also be reconsidered
Previously: A · B · C. After a meaningful change:
I may be ready to reconsider my whole system deeply after a strong new signal. But the examination itself reveals what has actually changed and what continues to work. With Fable and Astra, I was describing precisely that readiness and a process of thinking that had not yet reached its conclusion. Being able to return to the foundations does not require discarding everything that came before each time.
Towards the end, Zhenya recalled Carol Dweck’s Mindset and the idea that abilities can develop. Adding “yet” to “I cannot do this” helps him. The word keeps the possibility of learning open. He also connects learning with his own choice: he is willing to work through a new subject if it fits his direction. I like that interest and the freedom to defer a topic remain alongside confidence in our ability to learn.
Change can be difficult. A familiar way of working may stop matching what someone wants to do, as it did for Sasha; Zhenya may still be looking for a role that suits him. None of us knows what will happen in a year or two. Even now, though, we can notice a question that deserves reconsideration and give it attention at a pace that suits us.
Evgeny asks whether a new topic fits his chosen direction. If it fits, Evgeny keeps the possibility of learning open with ‘I cannot do this yet’. If it does not fit his direction, it can wait. Both possibilities remain visible.
New topic
Does it fit my direction?
I cannot do this yet
I can learn
It can wait
After this conversation, the choices look more concrete to me. An unread digest can be switched off. When a study text has lost its connection to the goal, it is worth asking why it is needed and moving to practice. I can assess a day filled with activity by which actions brought the needed result closer. I can give a thought that has not yet taken shape in words some time, then return and try to express it myself.
Sasha suggested starting with a question that feels odd in an interview and is useful to ask yourself: what do I want, and where do I see myself in a few years? The answer can change. What matters to me is retaining my own direction through each change, with time to live alongside it. Questions or disagreements with our approaches can be left under the episode. Those comments also help us gather topics for future conversations.
For working with AI on engineering tasks, I have put together an AI-assisted engineering programme. If you are choosing how to learn this kind of work, look at its contents and consider how they fit your own tasks.
Sources and Further Reading
- 1A conversation between Aleksei Litvinau, Alexander Polomodov and Evgeny Sergeev about attention, changing ways of working and personal limits.
- 2The official description of five stages for handling incoming tasks and material; Zhenya describes his own adaptation of these ideas.
- 3The book Zhenya named and an article about freeing attention; I adapted my own writing practice from these ideas.
- 4The tool in Zhenya's example; commands for saving a segment depend on the device.
- 5Karpathy's approach to source material, an interlinked collection of text notes and its maintenance with a model.
- 6An explanation of the technology radar's rings; Sasha uses a similar idea with his own categories of attention.
- 7A classic account of agents and their environments, which Sasha recalled when a familiar concept became an available working tool.
- 8The research example Sasha mentioned that combines program generation with automated evaluation.
- 9The source of the title used as a metaphor for human attention in the conversation; the paper itself concerns the Transformer architecture.
- 10Further reading on an experiment in returning to already-started creativity tasks after an intervening activity.
- 11Further reading on transitions between work tasks and continuing thoughts about unfinished work.
- 12A developer survey about demands and resources in the context of generative AI use; a study of associations, rather than a test of our personal practices.
- 13The book in Zhenya's account and the author's talk about the possibility of developing abilities.