Longreads
Deep dives into engineering practices, research, and architectural decisions — with room for evidence, context, and practical application.
Where to start
Choose a task: start with an explanation, then take the next step.
Assess the value of AI
Design an agent system
Understand model infrastructure
Improve how your team works
topics
- Preparation for Research Insights Made Simple episode 37ENFellow
AI and the data platform: connecting AI to the organization’s data
Four archetypes of data unification and eight descendant platforms after Nikolay Golov’s scheme, a separate platform option built on a PostgreSQL provider with DuckLake and CDC, plus metric definitions, permissions, freshness and verified agent actions.
- polomodov.tech · #AI · #Payments · #ProductEN
An Agent with the Right to Pay: Authority, Consent and Liability
How to delegate a purchase to an agent: AP2 v0.2 authority, consent, constraints and dispute evidence. Visa rules, proposed Amex protection, choice experiments and the limits of market forecasts.
- Curriculum comparisonENFellow
Distributed Systems Courses: Comparing Approaches
How MIT, Cambridge, UW, Yandex SDA, ITMO, and CMU teach distributed systems: theory, labs, and parallels with our course. The Maelstrom workshop and ideas for developing the syllabus.
- polomodov.tech · #AI4SDLC · #AgentsENFellow
The Boundaries of Real Autonomy: What Work an Agent Completes End to End
How to measure agent autonomy so that lab tests, comparative benchmarks and production agree: the unit of work, METR's time horizon, the drop in multitask environments, reliability versus capability, the law of the long chain, success thresholds by cost of failure, an autonomy scorecard, pseudo-autonomy, and handing control back to a human
- polomodov.tech · #AI4SDLC · #AgentsENFellow
The Agent Harness: How the Environment Around the Model Changes
A research article on the agent harness: its seven parts, how functions moved from prompt to model and from model to runtime across four eras, where the harness effect is measured and where it vanishes, which parts the labs are betting on, and eight dated forecasts to 2031. Plus a harness ledger: a register of assumptions about the model with a retest date
- Position ahead of the roundtableENCTO
How AI Changes the Tech Lead’s Role: A Position for the Roundtable
A position prepared ahead of the September 30 roundtable: architecture, RFCs, code review, autonomy boundaries, engineer development and a first-month plan. Based on talks and Book Cube posts.
- Independent researchENCTO
The AI-Native Organization: What It Is and How to Get There
A study of the AI-native organization: redesigning workflows, human and agent roles, a shared platform, measuring outcomes, and a practical transition plan. The author's notes, research, and industry experience.
- Preparing for 3 AImigoEN
Preparing the September 2026 AI Digest: Cheaper Generation, Costlier Verification
The news gathered for 3 AImigo episode seven: six September stories on cost per task, CI and code review, juniors and seniors, agent harnesses, control of agent actions, and the pacing dispute.
- Preparing for DotNext 2026ENFellow
IDE vNext: a workspace for engineers and their agents
Preparation for the DotNext roundtable: the engineer’s role, attention and parallel agents, the connection between IDEs and internal developer platforms, portability, traceability, and the cost of accepted work.
- polomodov.tech · #MLSystems · #SREENFellow
Operating LLM Inference: What to Promise and How to Serve Requests Under Overload
Why a throughput chart is not a promise, which three numbers an inference SLO is made of, why capacity is measured in memory for contexts, what the failure taxonomy looks like, and what to serve instead of a 500
- Research Insights Made Simple #30EN
Developer Productivity: Google’s Human-Centered View
A longread on Developer Productivity for Humans: developer goals, measurement, work environments, quality, teamwork, and AI tensions. Google research, Book Cube and Medium reviews, and a practical first-month plan.
- polomodov.tech · #AI · #Economics · #PlatformsEN
The agent economy: who benefits when machines make deals
How the AI agent economy is developing: real trades, task markets, delegation, payments, inequality, and the full cost of a verified outcome.
- polomodov.tech · #AI4SDLC · #SecurityENFellow
AI Security in Software Development: Where Agents Become the Risk and Where They Already Find the Bugs
Updated 19 September 2026: code and AI-agent security, injection in the wild, Plugin4Shell, secrets, Codex Security and Glasswing results, active EU CRA reporting and Russia’s FSTEC rules.
- polomodov.tech · #AI4SDLC · #ArchitectureENFellow
When Agents Write the Code: What Remains Engineering
How AI agents reshape software development: bottlenecks, architecture, outcome verification, autonomy, context and memory, junior training and a three-year outlook.
- polomodov.tech · #Data · #Product · #EconomicsENCTO
Where’s the Profit in Data, Lebowski? Three Routes from Data to Money
Data licensing, internal decisions and product features: economics, rights, Reddit, Avast, Statist, Hippo, Netflix and Uber. How to prove the impact and choose a route.
- polomodov.tech · #MLSystems · #PlatformEngineeringENFellow
The LLM Inference Frontier: Promised Gains and the Evidence Behind Them
Which model, cache and hardware changes promise large gains in LLM inference. Mechanisms, measurements and limits of 2026 research, from hybrid attention and FP4 to disaggregated computation and agent workloads.
- polomodov.tech · #AI4SDLC · #LeadershipENFellow
Minority Report: When Code Became Cheap — Where Do Value, Responsibility, and Expertise Live Now?
Positions for the Deep Tech Night panel, filed before it started: where the bottleneck moved, what an agent may do without a human and who is accountable, which expertise stays human and how to grow it, and which corporate AI bet is overrated
- polomodov.tech · #Leadership · #VentureENFellow
Where a Fellow-Level Technical Leader Can Have the Most Leverage: A 2026 Ranking of Venture Frontiers
Nine venture frontiers ranked for a Fellow-level technical leader on six criteria — from profile leverage to portability — plus a 90-day validation path.
- polomodov.tech · #AI · #RoboticsENFellow
Physical AI: How Machines Learned to See and Act — and Whether Robotics Has Had Its ChatGPT Moment
A history of Physical AI from cybernetics, Shakey, and DARPA to Waymo, VLA models, and world models: how robot engineering changed, why humanoids surged, and what still separates the field from a mass-market ChatGPT moment
- Code of Leadership · S2E13 · Book CubeEN
Subscriber AMA: Reading, Career, Side Projects, and AI Development
A long-form conversation about reading and personal knowledge systems, System Design Space, side projects, a career transition, reliable agentic development, AI economics, and cautious AGI/ASI scenarios — with a complete source map for the episode and its preparation.
- polomodov.tech · #MLSystems · #PlatformEngineeringENFellow
LLM Inference as a Distributed System: From KV Cache and vLLM to Phase Disaggregation
How LLM inference differs from conventional ML, where Orca, PagedAttention and vLLM came from, which techniques accelerate each layer, and how prefill/decode disaggregation emerged
- polomodov.tech · #AI4SDLC · #CareerENCTO
The Junior Engineer After Code: Growing Engineers When Agents Do the Implementation
Why agents can accelerate a first pull request without turning a junior into a mid-level engineer: a new learning model, growth signals, interviews, and AI apprenticeship for engineering teams
- 3 AImigo · S1E1 · #AI4SDLCENFellow
Where AI in Software Development Stands Now: Defaults, Delegation, and the Limits of Autonomy
An evidence-based AI4SDLC snapshot as of August 12, 2026: maturity across the lifecycle, shifting bottlenecks, changing engineering roles, four company operating models, and the transformation consulting market
- polomodov.tech · #LocalAI · #MLSystemsENFellow
Local LLM Inference in 2026: Models, Hardware, Performance, and Cost
Which open-weight models run efficiently on existing computers, one GPU, and 128 GB workstations; memory arithmetic, published tokens-per-second benchmarks, current prices, and use-case configurations
- polomodov.tech · #Leadership · #CareerENCTO
Your Last 90 Days at a Company Start Before You Resign: Understanding What You Want, Handing Over Responsibility, and Leaving Well
A practical model for closing a chapter: diagnosing dissatisfaction, reflecting on your career, testing internal mobility, searching in parallel, and handing over responsibility without gaps
- polomodov.tech · #Leadership · #CareerENCTO
The CTO’s First 90 Days Start Before Day One: Choosing the Role, Negotiating the Mandate, and Passing Probation
A practical transition model for technology executives: deep research, mutual due diligence, the commit point, a first-month map, system bets through day 90, and a red-flag protocol
- polomodov.tech · #AI4SDLC · #PlatformEngineeringENFellow
An End-to-End Player Without Its Own Silicon: Building AI Development in Russia from Traces and Harnesses, Not Pretraining
What to do when you have the people, the harness, your own serving capacity and your own traces, but no access to new accelerators: a compute-allocation doctrine, training under weak interconnect, serving efficiency, an honest account of open-weight dependency, and a rehearsed base-substitution drill
- polomodov.tech · #AI4SDLC · #PlatformEngineeringENFellow
AI Development as a Co-Evolving Stack: Hardware, Models, Harnesses, Tools, and Traces
Why the advantage in AI development comes not from one model but from the speed of the hardware → model → harness → tools → outcomes → evals loop, and which layers an enterprise should rent, adapt, or own
- polomodov.tech · #AI4SDLCENFellow
The Economics of AI Development: A Complete Analysis from Tokens to Accepted Work
Why cheaper models do not shrink the AI budget: cost per accepted task, trace budgets, FinOps, routing, vendor lock-in, contracts, and fully loaded local-model TCO
- polomodov.tech · #PlatformEngineering · #SREENFellow
Who Made Kubernetes So Complex? How Complexity Moved into Platforms and AI Agents
Why Kubernetes became boring but not simple: Ingress and Gateway API, correlated failures, the internal-platform paradox, and agents replacing kubectl
- polomodov.tech · #AI4SDLCENFellow
Agent Stack Configurations: A Complete Analysis of Eight Options
A full study of eight configurations covering cost structure, lock-in, and failure modes, with a boundary-based threat model, comparison matrices, a checklist, and a decision tree
- polomodov.tech · #Architecture · #AI4SDLCENFellow
AI for Software Architecture: Why the Architect Copilot Still Hasn’t Arrived
Inside a systematic review of 51 studies (Bucaioni et al., ACM TOSEM): where AI already helps architects, why it remains local and reactive, and what engineering foundation a living architecture practice requires
- polomodov.tech · #AI4SDLCENFellow
How to Evaluate AI Agents: From a Convincing Answer to a Reproducible Engineering Episode
Replayable episodes, hidden judges, a five-layer scorecard, and a scenario matrix for evaluating agents across the SDLC, architecture, and data platforms