Book Cube: 2026, page 2
Book Cube publications from 2026, page 2: bilingual notes, sources, and original Telegram posts.
- · #AI
A Day at DotNext: Keynote, 3 AImigo Live Stream, and IDE vNext
Alexander Polomodov's DotNext schedule: a keynote on the future of AI-assisted development, a 3 AImigo live stream, a discussion about IDEs for agents, and the afterparty.
- · #Inference
Breaking Down the vLLM Whitepaper: KV Cache, PagedAttention, and Fast Inference | Research Insights #32 (Series #Inference)
An announcement for Research Insights Made Simple #32 on September 28: vLLM, KV cache, PagedAttention, request scheduling, and state transfer between nodes.
- · #Management
A Team with AI: What Changes in a Manager's Work | Code of Leadership S2E22 with Dima Tverdokhlebov (Series #Management)
A live Code of Leadership S2E22 conversation with Dima Tverdokhlebov about moving from large companies to his own venture, delegating to people and agents, hiring, Omnius.team, and talant.club.
- · #AI4SDLC
SWE-bench: How to Choose an Agent When the Leaderboard Cannot Distinguish the Leaders (Series #AI4SDLC)
A discussion of Coding Agents Have Converged: what close SWE-bench scores mean, why adjacent ranks do not prove superiority, and how to compare agents on your own tasks.
- · #Management
How to Lead in a Field You Don't Yet Understand | Leonid Chernyy | Code of Leadership S2E23 (Series #Management)
A September 29 live conversation with Leonid Chernyy about changing fields, trusting experts, delegation, and two rules for leaders.
- · #Management
The live stream with Fedor Sukharev is starting
A short invitation to join the live stream with Fedor Sukharev and send questions.
- · #Management
Episode #30 Materials: Developer Productivity for Humans (Series #Management)
Materials from the solo episode of September 21, 2026: developer goals, metrics, quality, teamwork, and AI tensions.
- · #AI
Mixedbread: How to Organize Knowledge Work for an Agent (Series #AI)
Benjamin Clavié's talk on dividing an agent's research work, and how that model connects to Mixedbread's retrieval infrastructure and Toast 1.
- · #Management
Code, Cars, and Insurance Products: What a CTO Has to Relearn | Fedor Sukharev | Code of Leadership S2E21 (Series #Management)
A live conversation with Fedor Sukharev about moving between industries, automotive development at Arrival, QIC insurance products, AI, and earning a team's trust.
- · Книжный куб
Live stream on AI4SDLC and lessons learned
The AI4SDLC live stream on my lessons learned is starting; join us and ask questions.
- · #Books
Modern Computer Architecture and Organization — From the Processor to the AI Data Center (Books)
A look at Jim Ledin's first and third editions: from processors and memory to dedicated chapters on GPUs and LLM data-center architecture.
- · Книжный куб
Live stream on Developer Productivity
The Developer Productivity live stream is starting; join us and ask questions.
- · #Management
Research Insights Made Simple #30: Developer Productivity for Humans
September 21 at 17:00 Moscow time (UTC+3): Alexander Polomodov’s solo episode on developer productivity, covering goals, measurement, quality, teamwork, and the consequences of AI use.
- · #AI
Y Combinator: Hardware, Agents, and Founders (Series #AI)
A discussion of The State of Startups in 2026: demand for manufacturing in the US, the case against every product building its own harness, experienced founders, and starting without a co-founder.
- · #AI4SDLC
Jev: intelligence for an ordinary if statement (#AI4SDLC)
How TypeSafe turned Diogo Almeida’s ideas into Jev: small decisions for software, early external tests, and the limits of the promise of no hallucinations.
- · #AI4SDLC
Diogo Almeida: AI That Does Not Need Constant Supervision
A discussion of the InstructGPT coauthor's talk on why human approval, answer correctness, and reliable automation require different checks.
- · #AI
3 AImigo S1E6 Materials: How to Keep Up with AI Changes Without Draining Your Energy (Category #AI)
Video, audio, and a recap of 3 AImigo episode six: filtering AI updates, learning with agents, protecting time to think, and measuring progress beyond hours spent.
- · #Books
Regenerative Software — Chad Fowler (Category #Books)
Why cheaper code generation shifts engineering toward system boundaries, behavioral evaluations, and preserving decision rationale. A recommendation of Chad Fowler’s work in progress.
- · #AI4SDLC
3 AImigo S1E5 Materials: A Junior Without Easy Tasks (Category #AI4SDLC)
Video, audio, and a recap of 3 AImigo episode five: finding a first job in the AI era, personal projects, engineering fundamentals, and learning with mentors and agents.
- · #Architecture
IT as Lego: Why This False Metaphor Does More Harm Than Good—and What to Use Instead (#Architecture)
Why the Lego metaphor does more harm than good in IT systems, why the image of a living system is more useful, and how Chad Fowler develops this line in Regenerative Software.
- · #AI
Homa: Why GPUs Wait for the Network (#AI)
A breakdown of John Ousterhout's recent talk: how Homa reduces short-message latency, what the network benchmark actually shows, and what experiment an AI workload still needs.
- · #DistributedSystems
The Tail at Scale: How to Beat the Slow Tail (#DistributedSystems)
Dean and Barroso on latency tails: hedged and tied requests, small partitions, and controlled incompleteness, with connections to Monarch, Cassandra, QoS, and Harvest/Yield—and the limits of these techniques.
- · #AI
3 AImigo live: keeping up with AI changes without burning out
An invitation to the next 3 AImigo livestream on keeping up with the flow of AI changes while preserving your energy.
- · #AI4SDLC
When Code Became Cheap: Deep Tech Night Panel Materials (Category #AI4SDLC)
Recording and recap of the Deep Tech Night panel I joined: value from AI-assisted development, accountability for agents, junior education, and the cost of human attention.
- · #AI4SDLC
Research Insights Made Simple #31: AI4SDLC — what I would do differently
Announcing the director's cut of the Deep Tech Night talk: what to rent, adapt, and own in an AI stack. The episode comes out on September 22 at 13:00 MSK.
- · #Management
What Happens When Technical Debt Vanishes? (Category #Management)
A Google thought experiment asks how engineering work and productivity metrics change when one kind of technical debt disappears for good.
- · #PlatformEngineering
Kubernetes: too complex? Materials from DevOps Deflope No. 62 (Category #PlatformEngineering)
Materials from DevOps Deflope episode 62: Kubernetes complexity, internal platforms, and the trust boundaries for AI agents in a cluster.
- · #AI4SDLC
Developer Productivity for Humans: four tensions of AI
How AI shifts work between stages and people, affects debt and learning: a continuation of Google's developer productivity series and the authors' recommendations.
- · #Research
How I Read Whitepapers Now: Marginal Notes and AI
A personal note on reading research papers with marginal annotations and discussing them with Codex/Claude.
- · #AI4SDLC
How Vercel Built d0: An Agent's Evolution and Questions About Its Results
The evolution of Vercel's analytics agent d0, from a large prompt and specialist chain to a file-based environment, skills, and the eve framework—and the open questions about answer quality and accumulated knowledge.
- · #AI4SDLC
YC Paper Club: What Turns a Model into a Working Agent (Category #AI4SDLC)
Four engineering perspectives on agent harnesses: tools and memory, Prime Agent, local OpenJarvis, and the QM internal platform for dozens of personal agents.
- · #AI4SDLC
How AI Will Change Software Development: Organized Programming #92 Materials (Category #AI4SDLC)
Materials from a conversation with Kirill Mokevnin about how AI changes programmers’ work, team design, internal platforms, impact measurement, and engineering education.
- · #Leadership
Episode Materials: CTO Freedom in Startups and Corporations with Kirill Evseenko (Category #Leadership)
Recording and recap of Code of Leadership S2E20 with Zvuk CTO Kirill Evseenko: scaling START, technology economics, influence through managers, and engineering careers.
- · #Museum
Science Museum Souvenir Book (Category #Museum)
The London Science Museum souvenir book echoes the exhibition's logic: domains, chronology, and inventions represented in the collection.
- · #Data
Where’s the Data Profit, Lebowski? Episode One Materials (Category #Data)
Recording, companion article, and recap of the first episode with Andrey Tsybin, Nikolay Golov, and Alexander Polomodov: who buys data, what they pay for, and what remains after costs.
- · #AI4SDLC
Habitat: the evolution of OpenAI’s storage layer (Category #AI4SDLC)
Habitat’s evolution from a library to a Python service and Rust: operational control, a constrained API, analytics isolation, and latency under load.
- · #AI4SDLC
Codex: Why Cheaper Rewrites Still Need Architecture (Category #AI4SDLC)
Tibo Sottiaux's path from DeepMind to Codex: how agents make changes cheaper and why agreement on system boundaries becomes more important.
- · #Leadership
Code of Leadership: the Livestream with Kirill Evseenko Is Live
An invitation to a livestream with Zvuk CTO Kirill Evseenko about a CTO's career and freedom in startups and corporations.
- · #Research
[2/2] DistServe: Parallelism, Queues, and GPU Balancing (Category #Research)
The second DistServe breakdown: tensor and pipeline parallelism, queues, workload simulation, and balancing GPUs for a model, request profile, and SLO.
- · #Research
[1/2] DistServe: Why Separate Prefill and Decode (Category #Research)
A breakdown of DistServe: separate GPU pools for prefill and decode, KV-cache transfer, topology, and goodput under specified SLOs.
- · #AI
3 AImigo: the episode about junior engineers is live
A short update that the 3 AImigo episode about junior engineers is live.
- · #Leadership
Episode materials: the first 90 days as CTO
Materials for solo Code of Leadership episode 77: choosing the role, agreeing on authority, first-month diagnosis, and initial changes. Recording, slides, longread, and recap.
- · #Management
Randy Shoup on eBay: Engineering Sped Up, the System Did Not
A breakdown of Randy Shoup's LeadDev London 2026 talk: how, according to the speaker, eBay doubled engineering productivity with a standard DevOps playbook yet failed to change centralized planning, the feature factory, and what he describes as a pathological culture.
- · #AI4SDLC
AI4SDLC at IT Picnic: Talk Materials (Category #AI4SDLC)
Slides, recording and an edited recap of Alexander Polomodov’s final talk representing T-Bank, taking stock of AI4SDLC as he had developed it.
- · #Architecture
ArchDays in the AI Era (Category #Architecture)
How architecture conferences help rethink engineering processes and the architect's role in the AI era.
- · #Leadership
Episode Materials: Product Engineering with Gleb Mikheev (Category #Leadership)
Recording and recap of Code of Leadership episode 76: product ownership, new constraints as development accelerates, and engineering growth when working with agents.
- · #AI
Stanford MS&E435: From Megawatts to Molecules — the Complete Course in Nine Reviews
A closing post on Stanford MS&E435: why the course offers a useful map of AI economics, with links to all nine reviews—from GPUs and data centers to coding agents and drug discovery.
- · #AI
3 AImigo S1E5: A Junior Engineer Without Simple Tasks — How Do You Enter IT in the AI Era?
A preview of the fifth 3 AImigo episode on how an aspiring engineer can enter the profession when AI agents are taking over traditional entry-level tasks.
- · #Management
A CTO's First 90 Days: Extended Livestream
An invitation to an extended version of the talk on a CTO's first 90 days, followed by a live Q&A.
- · #AI
PyTorch as a Portability Layer: Why Cambricon and Ant Group Joined
Why Cambricon and Ant Group joining the PyTorch Foundation is not a purchase of control over the code, but an attempt to reduce the software cost of moving away from CUDA—and which technical signals will reveal whether it works.