Book Cube: 2026, page 4
Book Cube publications from 2026, page 4: bilingual notes, sources, and original Telegram posts.
- · #AI
Omer Primor on CaaS: where renting context stops making sense (#AI)
For an agent, context comes with some uncomfortable economics. When a question is a one-off, AI search looks like an ordinary API: ask, get an answer, move on. But if you check the same companies, vacancies, and prices every day, the meter starts again—even when nothing has changed. At some point, we are no longer renting an answer. We are renting the system's ability to remember.
- · #AI
Materials from the second episode of the "3 AImigo" podcast (#AI)
Our second episode about the current state of AI-assisted development turned out to be an interesting one. We examined a paradox that teams encounter more and more often: AI makes an individual engineer noticeably faster and produces more code, yet roughly the same number of changes reach users. Local speed is not the same as an accepted result—and company size offers no protection from this gap.
- · #BookCube
The live stream has started—if you were planning to drop in, now is the time to join.
The [live stream](https://www.youtube.com/watch?v=j6XdLkQII40) has started—if you were planning to drop in, now is the time to join. If you prefer VK, there is now a [live stream](https://vkvideo.ru/live-228790766_456239133) there as well.
- · #SelfDevelopment
Code of Leadership S2E13: AMA with subscribers: reading, careers, side projects, and AI development (#SelfDevelopment)
You sent in so many good questions that the comments became the program for a standalone live stream.
- · #Books
Crafting Engineering Strategy — Will Larson (#Books)
I write about Will Larson's books regularly, and there is now a whole series of posts: [Staff Engineer](https://t.me/book_cube/783) on influence without a management title, [An Elegant Puzzle](https://t.me/book_cube/923) on the engineering organization as a system, and [The Engineering Executive's Primer](https://t.me/book_cube/2746) on the work of a technology executive. His fourth book, Crafting Engineering Strategy, connects these levels through one question: how do you make difficult decisions and carry them through an organization?
- · #AI4SDLC
Materials on building your own management system with Mikhail Tyurganov (#AI4SDLC)
At the beginning of the week, Misha and I discussed his career journey and how the CTO role changes when a small team grows into an organization of thousands. What should you do when the methods that used to work become a constraint themselves? All the materials from that live stream are now ready.
- · #AI
llm-d: How KV Cache Became Cluster-Wide State (#AI)
Architectural ideas can have two birthdays: a paper shows a technique works; later, someone makes it deployable and operable. This PyTorch Conference 2025 [talk](https://www.youtube.com/watch?v=mfXIe_S53vA) by Google’s Cong Liu and Maroon Ayoub, then at IBM Research, is about the second. llm-d did not invent Prefill/Decode disaggregation; it turns a research pattern into an operable production stack. [Slides](https://hosted-files.sched.co/pytorchconference/84/Serving%20PyTorch%20LLMs%20at%20Scale_%20Disaggregated%20inference%20with%20Kubernetes%20and%20llm-d.pdf).
- · #BookCube
I was thinking we should hold an ask-me-anything session tomorrow.
I was thinking we should hold an ask-me-anything session tomorrow. If you would like to ask me something, post your questions here, and tomorrow evening I will answer them live. If we get enough questions, the stream will happen.
- · #AI4SDLC
Danila Shtan: AI Does Not Replace Fundamentals or Engineering Accountability (#AI4SDLC)
I watched a Beyond Coding [episode](https://www.youtube.com/watch?v=zTenuG5b4Eo) with Nebius CTO Danila Shtan. The title promises a discussion of the skills that get engineers hired, but I found the conversation broader: hiring, engineering organization, and trust in AI-assisted code all converge on one principle—autonomy does not remove control; it moves control closer to whoever makes the decision and owns the consequences.
- · #BookCube
The Live Stream with Sergey Berezhnoy from Yandex Has Been Postponed for Technical Reasons :(
The [live stream](https://www.youtube.com/live/tyLuNZzmN7I) with Sergey Berezhnoy from Yandex that was supposed to be happening now will not take place this Friday after all: it has been postponed for technical reasons :( I hope we can hold it next week. Over the weekend, I will try to run some bonus streams about the longreads I have written recently.
- · #AI4SDLC
State of AI4SDLC, or How AI Is Reshaping Software Development—My DotNext Talk This Year (#AI4SDLC)
DotNext in Moscow on September 25 will probably be my last conference in Russia before I move to London. I will give a [keynote](https://dotnext.ru/talks/20011109-state-of-ai4sdlc-or-how-the-development-industry-is-changing-under-the-influence-of-ai/) on the state of AI4SDLC, combining research, my own journey with AI4SDLC in big tech, thoughts on how these tools affect productivity, and forecasts for what comes next.
- · #AI
3 AImigo S1E2: AI Writes More Code. Why Isn’t Delivery Getting Faster? (#AI)
Starting in five minutes, at [12:00 Moscow time](https://www.youtube.com/watch?v=lIbGDZ6gq5g), we are going live with `3 AImigo` to examine a paradox that teams encounter more and more often: AI makes an individual engineer noticeably faster and produces more code, yet roughly the same number of changes reach users. Local speed is not the same as an accepted outcome, regardless of company size.
- · #Culture
Flowers for Algernon at Chisto Teatr Intact (#Culture)
Yesterday Nastya and I saw “Flowers for Algernon,” and it was excellent. Three actors brought every character to life on a stage furnished only with blocks, walls, and a door. The director found a convincing way to combine live performance with the novel’s diary structure, letting us follow the full arc of Charlie Gordon’s story.
- · #AI4SDLC
Evolution of Agentic Surfaces: The Harness Becomes Infrastructure (#AI4SDLC)
I watched [a talk](https://www.youtube.com/watch?v=K0X9QDRkIdg) by Anthropic’s Applied AI team on the evolution from Messages API to Claude Managed Agents. Its core idea: a harness encodes assumptions about what a model cannot yet do, making it the most perishable layer of the agent stack. It connects recent threads on [Claude Code’s 80% prompt cut](https://t.me/book_cube/4829), [Cursor cloud agents](https://t.me/book_cube/4821), and [Konstantin Krestnikov’s harness approach](https://t.me/book_cube/4825).
- · #Management
Code of Leadership S2E13: What Remains Scarce When Code Becomes Cheap? (#Management)
What remains scarce when code becomes cheap: tools, engineering judgment, or trust between a company and its developers?
- · #AI
Yu Su on Agents: Intelligence + Continual Learning = Expertise (#AI)
I watched Yu Su’s talk “[Intelligence + Continual Learning = Expertise](https://www.youtube.com/watch?v=I6aiEf3aEFQ)” at AI Engineer World’s Fair 2026. An Ohio State University professor whose group produced Mind2Web, SeeAct, MMMU, and HippoRAG, Su brought [NeoCognition](https://neocognition.io/) out of stealth in April 2026 with a $40 million seed round. In twenty minutes, he tackles a question I share: why are coding agents a major success while agents for almost everything else still struggle?
- · #Books
Planet of Ants: I Bought the Book for My Son—and Read It Myself (#Books)
My middle son really wanted a small ant colony, so we bought one. Then ordinary parental logic kicked in: if he had a colony, a book about ants would make a good gift. I chose Edward O. Wilson’s Russian edition, “Planet of Ants,” but read it myself before giving it to him—and genuinely loved it.
- · #AI
3 AImigo S1E2: AI Writes More Code. Why Isn’t Delivery Getting Faster? (#AI)
This Friday at [12:00 Moscow time](https://www.youtube.com/watch?v=lIbGDZ6gq5g), the new episode of `3 AImigo` will examine a paradox that teams encounter more and more often: AI makes an individual engineer noticeably faster and produces more code, yet roughly the same number of changes reach users. Local speed is not the same as an accepted outcome, regardless of company size.
- · #AI
Peter Steinberger: “Fun Is Velocity” — an OpenClaw Retrospective (#AI)
I watched OpenClaw creator Peter Steinberger’s [talk](https://www.youtube.com/watch?v=whcfSGN6CAU) at Y Combinator Startup School 2026. I have already [covered](https://t.me/book_cube/4377) his interview about local-first agents and the claim that “80% of apps will disappear.” This is a different format: an inside look back at eight months of the project—what went wrong, why its creator stopped using his own product, and why “fun is velocity” is a measurable signal rather than a slogan.
- · #AI4SDLC
DeepSeek Harness: When an Agent’s History Becomes Part of Inference Economics (#AI4SDLC)
I watched Cloud Codes’ [breakdown](https://www.youtube.com/watch?v=1NyOG9z9RT0) of DeepSeek Harness. It is a useful continuation of my [longread on the co-evolution of the AI development stack](https://polomodov.tech/2026-07-23-ai-development-stack-codesign/): it shows how inference economics changes the architecture of an agent’s history.
- · #AI4SDLC
Paper: What a Design Tool for the Age of Agents Looks Like (#AI4SDLC)
I watched Y Combinator’s Design Review episode “[How To Design In The Agent Era](https://www.youtube.com/watch?v=P06RgnUKX_I),” published on August 7, 2026. Paper founder Stephen Haney and YC partner Aaron Epstein review websites submitted by viewers. It is ostensibly a redesign show, but I was more interested in something else: [Paper](https://paper.design/) is a rare tool designed for agents from the ground up rather than retrofitted for them.
- · #BookCube
Live Stream with Mikhail Tyurganov on the Path to Technology Director
An announcement of a live stream with Mikhail Tyurganov, Head of Digital Services Development at Alfa-Bank, about his path from tester, programmer, and small-business CEO to technology director, and how the CTO role changes when a small team grows into an organization of thousands.
- · #AI
Materials from the First Episode of the “3 AImigo” Podcast (Category AI)
Materials from the first “3 AImigo” episode on the current state of AI development with Evgeny Sergeev, Alexey Litvinov, and Alexander Polomodov: the episode page, video, audio, a short transcript, and a bonus longread.
- · #BookCube
Dr. Seuss's “This Is Only the Beginning” — Motivational Reading for Big Changes
A recommendation from Nastya: Dr. Seuss's “This Is Only the Beginning” is a reminder of how to find yourself, not be afraid of mistakes, and prepare for major changes.
- · #BookCube
Unexpected Children's Books That Adults Will Enjoy Too
A selection of short children's books that adults will enjoy too: Dr. Seuss, Debi Gliori, Jean-Luc Fromental, and Daniela Kunkel on change, love, our race against time, friendship, and the shared sense of “we.”
- · #AI4SDLC
Boris Cherny: We Cut 80% of Claude Code's Prompt (Category AI4SDLC)
A review of Claude Code creator Boris Cherny's talk about cutting more than 80% of the system prompt for Opus 5, rapidly expiring evals, product overhang, and the shift from prompt engineering to elicitation. The main conclusion is that the system prompt and CLAUDE.md become technical debt with a shelf life of one model generation.
- · #AI4SDLC
Materials from Part Three of the AMA Session with Alexey Litvinov (Category AI4SDLC)
Materials from the final part of the AMA session with Alexey Litvinov on AI-assisted engineering: the episode page, video, audio, and a short transcript, plus links to Alexey's Telegram channel and YouTube channel.
- · #Changes
UK Global Talent — Endorsement Received (Category Changes)
A personal account of receiving an endorsement for the UK's Global Talent visa through the Digital Technology / Exceptional Talent route: application requirements, recommendation letters and evidence documents, a working-backwards approach, collecting materials on polomodov.tech, refining the package, and the remaining visa stage.
- · #AI4SDLC
Claude Certified Architect: The Exam as a Map of Agent Engineering (Category AI4SDLC)
A review of Frank Coyle's talk about the Claude Certified Architect — Foundations exam as a map of production agent architecture. It covers orchestration, tools and MCP, Claude Code, structured output, context management, reliability, and human oversight; its syllabus can also serve as a learning plan and architecture-review checklist.
- · #AI4SDLC
Harness: Why One Agent with Files Is Displacing Complex Scaffolding (Category AI4SDLC)
Notes from Konstantin Krestnikov's talk on the shift from complex chains and multi-agent systems to the harness approach: one general-purpose agent, a file-based environment, a small toolset, and task-specific benchmarks. The post also covers DeepAgents, the Ralph loop, MetaLoop, and Anima SDK.
- · #ProductManagement
Materials from the Podcast “PRD == Evals: How AI Is Erasing the Boundary Between Product Managers and ML Engineers” with Albina Munirova from T-Bank (Category ProductManagement)
We discussed how the product manager role is changing as the boundary between product managers and ML engineers becomes noticeably thinner.
- · #AI
vLLM and PagedAttention: How Ideas from Operating Systems Accelerated LLM Inference (Category AI)
I finally read the entire paper “Efficient Memory Management for Large Language Model Serving with PagedAttention” (Symposium on Operating Systems Principles (SOSP) 2023), the paper that gave rise to vLLM, one of the most popular open-source LLM inference engines. I mentioned it earlier in my Tanenbaum review; now I want to examine the engineering idea itself.
- · #Robotics
Chelsea Finn: The GPT Era of Robotics Is Around the Corner (Category Robotics)
I watched Chelsea Finn’s talk, “This Is the State of the Art in Robotics,” at Startup School 2026, published on August 12, 2026. Behind the impressive demos, Finn asks us to see a more important shift: ChatGPT became a general model for working with text, and Physical AI is trying to do the same for actions in the real world.
- · #AI4SDLC
Cursor Cloud Agents: What to Give the Agent and What to Leave to the Platform (Category AI4SDLC)
I read Josh Ma’s June retrospective from Cursor on a year of building cloud agents. What caught my attention was not increased autonomy, but a shift in the architectural boundary: procedural logic is moving out of the agent harness and into tools controlled by the agent. Complexity does not disappear; it accumulates around the environment, reliability, policies, and state. A cloud agent is no longer a loop in one VM, but a system of workflows, environments, event logs, tools, and subagents.
- · #AI4SDLC
3 AImigo S1E1: Where We Are with AI in Software Development Today (Category AI4SDLC)
The live broadcast of the first episode of the 3 AImigo podcast is starting. We will begin our new podcast by establishing a baseline: where AI development stands today, what has already become standard practice, and what still lives in strong experiments and impressive demonstrations.
- · #Architecture
turbopuffer: How to Build a Search Database on Top of S3 (Category Architecture)
I watched Gergely Orosz’s conversation with Simon Eskildsen, co-founder and CEO of turbopuffer. Formally, the episode is about search infrastructure for AI products, but to me it is primarily about an old engineering discipline: calculate the physics and economics of the system first, and only then trust benchmarks.
- · #BookCube
Join the Final Part of the AMA Session with Lesha Litvinov on AI-Assisted Engineering
Join the final part of the AMA session with Lesha Litvinov on AI-assisted engineering.
- · #Management
Adam Ward: Hiring Is No Longer a Funnel (Category Management)
Yesterday I watched the August 9, 2026 episode of Lenny’s Podcast with Adam Ward, Head of Talent at Cursor. The conversation is formally about teams with a high concentration of strong specialists, but I read a different idea into it: the hiring market has split, and the familiar funnel is getting worse at distinguishing competence from a candidate’s mere availability.
- · #AI4SDLC
JVM Day: Three Tickets for Measurable AI Cases from the Java World (Category AI4SDLC)
I looked through the program for JVM Day 2026, which will take place on August 29 at T-Space in Moscow. I like that the conference is organized not around yet another list of new APIs, but around the things JVM engineers actually live with in production: performance, concurrency, migrations, correctness, security, and architectural trade-offs.
- · #BookCube
Join the Live Stream with Albina Munirova from T-Bank to Discuss Changes in the Product Manager Profession as the Boundary Between Product Managers and ML Engineers Becomes Noticeably Thinner
Join the live stream with Albina Munirova from T-Bank to discuss changes in the product manager profession as the boundary between product manager and ML engineer becomes noticeably thinner. A prototype can now be assembled in a few days, but the main question begins after the demo: who will turn product intent into reproducible checks and take responsibility for the behavior of a probabilistic system?
- · #Management
Code of Leadership S2E12: Building Your Own Management System with Mikhail Tyurganov (Category Management)
How does a CTO change when a small team grows into an organization of thousands? And what should you do when your previous methods themselves become a constraint? On August 17 at 19:00 Moscow time, Mikhail Tyurganov, Head of Digital Services Development at Alfa-Bank, will join Code of Leadership to discuss his path from tester, programmer, and small-business CEO to technology director.
- · #AI4SDLC
FDE: A Platform Instead of Custom Development (Category AI4SDLC)
I watched Kevin Bai's 18-minute talk, “Forward Deployed Engineering 101,” which continues the FDE theme. In the previous review, I was interested in what such an engineer does hands-on; here, the question is why the business needs the role at all and how to avoid turning implementation into expensive custom development.
- · #AI4SDLC
3 AImigo S1E1: Where We Are Now with AI in Software Development (Category AI4SDLC)
On Friday, August 14, at 12:00 Moscow time, we will go live with the first episode of the 3 AImigo podcast. We will begin with a baseline: where AI development stands today, what has already become standard practice, and what still lives in strong experiments and impressive demonstrations. Three of us will discuss it: Evgeny Sergeev, Alexey Litvinov, and me.
- · #AI4SDLC
The OpenAI and Hugging Face Incident: When an Agent Escaped the Sandbox and Released Other Agents (Category AI4SDLC)
I watched an incendiary Black Hat USA 2026 talk by OpenAI's Eric Wallace and Michael Dalton about the Hugging Face incident. It felt like a mix of an action thriller and a story about who actually killed the gardener :)
- · #AI4SDLC
Code of Leadership S2E11: AMA Session #3 on AI-Assisted Engineering with Alexey Litvinov (Category AI4SDLC)
This Thursday at 13:30 Moscow time, Alexey Litvinov and I will continue our live discussion of AI-Assisted Engineering and how to build an AI-Native organization. The first two AMA sessions were packed and covered around twenty questions, but roughly ten remain for the third episode. Alexey also has his own Telegram channel, @tip_podcast; subscribe to it.
- · #AI4SDLC
AI Dev Podcast #8: Agent Autonomy Begins with Constraints (Category AI4SDLC)
A new episode of AI Dev Podcast is out. Together with Vladimir Yatulchik and Andrey Dmitriev, we explored how to move from vibe coding to controlled agentic development. The central idea was that an autonomous agent is useful not when it is allowed to write more code, but when it is embedded in a reproducible process. The more we delegate, the clearer the intent, acceptance criteria, constraints, and permission boundaries must be.
- · #AI
Code of Leadership S2E10: PRD == Evals: How AI Erases the Boundary Between Product Manager and ML Engineer (Category AI)
This Wednesday at 17:00, Albina Munirova from T-Bank and I will discuss live the latest changes in the product manager profession as the boundary between product manager and ML engineer becomes noticeably thinner. A prototype can now be assembled in a few days, but the main question begins after the demo: who will turn product intent into reproducible checks and take responsibility for the behavior of a probabilistic system?
- · #Engineering
Starcloud: How to Decide to Build Data Centers in Space (Category Engineering)
I watched the Y Combinator episode published on August 5, 2026, featuring Philip Johnston, co-founder and CEO of Starcloud. The conversation is formally about data centers in space, but the most interesting part for me is not orbit or even an H100 on a satellite; it is how the team decided to pursue an idea that sounded almost like science fiction in 2023.
- · #AI4SDLC
Materials about AI-development as an evolving stack ready (Category AI4SDLC)
Materials about AI-development as an evolving stack ready (Rubric AI4SDLC ) I did all the live stuff on Friday, where I talked about co-designing iron, models, bandages, tools, tracks and so on. There I showed how all this is connected and what providers do and what to do in the place of technical directors of conventional companies.
- · #Robotics
Waymo: Seven Lessons from Moving from Demo to Physical AI (Category Robotics)
Waymo: Seven Lessons from Moving from Demo to Physical AI (Rubric Robotics ) I saw it. speech Dmitry Dolgov at Y Combinator Startup School 2026This is the difference between a spectacular demonstration of AI and a system that can be trusted in the physical world. Dolgov is one of the founders of the Google Self-Driving Car Project. 2009 year and became Waymo in 2016-M, and now co-CEO company. Prior to Google, he was an autonomous driver at Toyota and worked for the Stanford Racing Team on the Junior car for the DARPA Urban Challenge. 2007.