Longreads
Deep dives into engineering practices, research, and architectural decisions — with room for evidence, context, and practical application.
- 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