Research Insights Made Simple

SGLang: Executing Structured Language Model Programs

SGLang: Efficient Execution of Structured Language Model Programs

A review of the SGLang paper: a language for LM programs, RadixAttention with a radix-tree KV cache, a compressed finite state machine for schema-constrained decoding, and up to 6.4× throughput without changing the model.

Long-running Agents: Work Handoffs

Effective harnesses for long-running agents

How an agent resumes work after a context reset: environment setup, a feature list, progress notes, Git, and browser verification.

SWE-agent: Interfaces Change Outcomes

SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering

A solo SWE-agent review: search, editing, validation and context, SWE-bench ablations, and the shift from a specialized interface to mini-SWE-agent.

vLLM and PagedAttention: Virtual Memory for LLM Serving

Efficient Memory Management for Large Language Model Serving with PagedAttention

A review of the vLLM and PagedAttention paper: why LLM serving is bound by KV cache memory, how paged virtual memory from operating systems removed reservation and fragmentation, and how that bought 2-4× throughput without changing the model.

AI4SDLC: What I Would Do Differently

What to rent and what to own in an AI stack: agent loops, context, permissions, tools, and verifiable engineering tasks. The extended Deep Tech Night talk.

Developer Productivity: Google’s Human-Centered View

Developer Productivity for Humans

Developer Productivity for Humans: goals, measurement, builds, onboarding, quality, teamwork, and AI tensions through Google research and Alexander Polomodov’s reviews.

AI Security in Development: The Agent Became an Actor

Срез по безопасности AI в разработке: где агент становится риском и где он уже ловит уязвимости

Updated September 19, 2026: code and AI-agent security, Plugin4Shell, in-the-wild injections, four Anthropic incidents, validating findings and fixes, CRA deadlines and draft FSTEC requirements.

AI-Native SDLC: Code Accelerated, Delivery Didn't

The AI-Native SDLC Playbook

A review of the AI-Native SDLC Playbook: redesigning planning, design, build, test, deployment, and maintenance around AI agents, versioned artifacts, and human control points. Guest: Anton Kosterin.

The Data Platform in 2026: From DWH to Lakehouse and AI Agents

How to separate OLTP from OLAP, where MPP warehouses reach their limits, what makes a lakehouse, why legacy migration takes years, and how AI agents change data-platform requirements. Guests: Nikolay Golov, Alexander Filatov.

AI Development as a Co-Evolving Stack

AI Development as a Co-Evolving Stack: Hardware, Models, Harnesses, Tools, and Traces

How hardware, models, harnesses, tools, traces, and evals form a co-evolving AI-development stack, and where enterprises should draw the rent, adapt, and own boundary.

Building Evals That Work

Production-grade evals for AI agents

How to turn AI-agent evals from one-off answer grading into a reproducible engineering system with replayable episodes, hidden judges, traces, scorecards, and release gates. Guest: Evgeny Sergeev.

Modeling Reliability from a Dependency Graph

Model Discovery and Graph Simulation: A Lightweight Gateway to Chaos Engineering

How to derive a dependency graph from traces, estimate availability with Monte Carlo, and prioritize chaos experiments. Guest: Anatoly Krasnovsky.

Why the Architect AI Copilot Still Hasn’t Arrived

Artificial Intelligence Support for Software Architecture Practice

What AI can do in software architecture, where context breaks, and why humans remain accountable for architectural trade-offs. Guest: Sergey Baranov.

The Economics of AI Development: From Tokens to Accepted Work

Why cheaper tokens do not guarantee a smaller AI budget, and how to manage cost per accepted task, trace budgets, routing, and local-model TCO.

How to Build a Governed Agent Stack

Конфигурации агентного стека: обвязка × модель × инструменты

How to choose a harness, model, and tools, define data and authority boundaries, and turn an agent stack into a governed enterprise system. Guest: Mikhail Trifonov.

Measuring Coding Agents in Dialogue

SWE-Together × SWE-INTERACT

A review of SWE-Together and SWE-INTERACT with Aleksey Litvinov: measuring coding agents when requirements emerge during the work. Guest: Aleksey Litvinov.

Why Coding Agents Do Not Eliminate Expertise

Agentic Coding and Persistent Returns to Expertise

A review of Anthropic's research with Evgeny Sergeev: why coding agents amplify domain expertise rather than eliminate it. Guest: Evgeny Sergeev.

Loop Engineering: Designing Systems That Run Coding Agents

Loop Engineering: The Anthropic Playbook for Designing Systems That Prompt Your Agents

A review of Loop Engineering with Maxim Smirnov: agent loops, generator/evaluator, hidden debts, and safe autonomous-system design. Guest: Maxim Smirnov.

AI in SDLC Report by IT One and Skolkovo

AI in SDLC Report

A review of the AI in SDLC report by IT One and Skolkovo: AI adoption maturity, effect metrics and practical AI4SDLC questions. Guest: Dmitry Nemov.

Early-2025 AI and Experienced Developer Productivity

Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

A review of research on early-2025 AI and experienced open-source developer productivity: methodology, results and limitations. Guest: Artem Aryutkin.

Impact of Generative AI in Software Development

Impact of Generative AI in Software Development

A review of Impact of Generative AI in Software Development: speed, review, quality, context and organizational conditions for AI adoption. Guest: Igor Kurochkin.

What Goes Around Comes Around... and Around, Part II

What Goes Around Comes Around... and Around

Part two of What Goes Around Comes Around... and Around: recurring trade-offs in data systems and modern databases. Guest: Alexey Svetlichny.

What Goes Around Comes Around... and Around, Part I

What Goes Around Comes Around... and Around

Part one of the review of What Goes Around Comes Around... and Around: historical cycles in databases and recurring architectural ideas. Guest: Alexey Svetlichny.

DORA Methodology

DORA Methodology

A review of DORA Methodology: the origin of DORA metrics, research limitations and careful organizational use. Guest: Igor Kurochkin.

Measuring Developer Productivity With DX Core 4

Measuring Developer Productivity With the DX Core 4

A review of DX Core 4: delivery, feedback, cognitive load and flow as a compact developer productivity measurement model. Guest: Evgeny Sergeev.

Measuring AI Code Assistants and Agents

Measuring AI Code Assistants and Agents

A review of Measuring AI Code Assistants and Agents: evals, change quality, review load and AI's impact on the whole SDLC. Guest: Evgeny Sergeev.

Measuring Developer Experience With a Longitudinal Survey

Measuring Developer Experience With a Longitudinal Survey

A review of Measuring Developer Experience With a Longitudinal Survey: measuring DevEx over time and linking surveys to process improvements. Guest: Artem Aryutkin.

What Do Developers Want From AI?

What Do Developers Want From AI?

A review of What Do Developers Want From AI?: real developer expectations for AI tools, trust, context and everyday work. Guest: Nikolay Bushkov.

Measuring Developer Goals

Measuring Developer Goals

A review of Measuring Developer Goals: developer goals, work context, interruptions and the link between DevEx and real effectiveness. Guest: Alexander Kusurgashev.

Architecture Governance

Architecture Governance

An interview on architecture governance: principles, ADRs, standards, exceptions and architectural coherence. Guest: Pavel Lakosnikov.

Data Platforms

Data Platforms

An interview about data platforms: self-service, ownership, data quality and platform thinking for data. Guest: Nikolay Golov.

Developer Productivity: DORA Metrics, SPACE, DevEx

Developer Productivity: DORA Metrics, SPACE, DevEx

A developer productivity episode on DORA, SPACE and DevEx: how to measure engineering effectiveness without turning metrics into team rankings. Guest: Artem Aryutkin.

AI-Enhanced API Design

AI-Enhanced API Design

A review of AI-Enhanced API Design: where AI helps with API design, how to validate results and why responsibility stays with engineers. Guest: Pavel Karavashkin.

Secure by Design at Google

Secure by Design at Google

A review of Secure by Design at Google: secure defaults, platform constraints and engineering practices that make security part of design. Guest: Artem Merets.

Defining, Measuring and Managing Technical Debt

Defining, Measuring and Managing Technical Debt

A paper review on technical debt: definitions, measurement, risk management and the connection between debt and delivery speed. Guest: Dmitry Gaevsky.

API Governance at Scale

API Governance at Scale

A review of API Governance at Scale: API catalogs, ownership, design reviews and rules that help large organizations evolve APIs without chaos. Guest: Daniil Kuleshov.