Aleksandr Furmanov
Aleksandr Furmanov is an individual software engineer and technical writer focused on AI agents, distributed systems, and the craft of building software. His public work emphasizes turning complex systems into understandable, maintainable software and sharing practical architecture lessons.
At a Glance
- Software engineers using Claude Code
- Teams operating multi-agent or skill-based AI coding workflows
- Developers building software factories and wanting observability into agent orchestration
- Maintainers debugging token usage, latency, and context behavior in agent harnesses
AI Tools by Aleksandr Furmanov
(1)harness-profiler
Claude Code Profiler for Agents
Discussions
No discussions yet
Be the first to start a discussion about Aleksandr Furmanov
Latest News
harness-profiler v0.2.2 released with correct running-state display for unfinished background calls
harness-profiler v0.2.1 released with improved background-call handback reporting, local profile opening, plugin restructuring, and a privacy page
harness-profiler v0.1.2 launched as the first public release
Published What Did My Software Factory Just Do?, introducing harness-profiler and explaining the motivation for profiling AI-agent workflows
Products & Services
An MIT-licensed, open-source Claude Code plugin/CLI that records skills, agents, subagents, and tool calls through hooks, then renders a terminal call tree and self-contained UML-inspired sequence-diagram page. It reports timing, context relationships, failures, and new/cached/output token counts; it supports Claude Code on a local machine and on the web, requires Node 18+ and git, and is not yet supported for Codex.
Market Position
harness-profiler is positioned as a specialized profiler for Claude Code agent harnesses rather than a generic application profiler: it reconstructs skills, agents, forked contexts, tool calls, timings, and token costs as a call tree and sequence diagram. Furmanov's design notes distinguish it from timeline/flame-chart approaches and say Perfetto was rejected because it sends readers to a third-party site; the tool instead produces one self-contained local page.
Founding Story
This is a personal engineering and writing practice rather than a company. Furmanov says he writes when a problem, surprising pattern, or architecture decision is worth documenting; in 2026 he built harness-profiler after his local Claude Code skills and agents grew into a software factory whose behavior became difficult to track.
Business Model
Revenue Model
Furmanov's public project is an open-source, MIT-licensed tool distributed from GitHub and the Claude Code plugin marketplace; the reviewed sources describe no paid plans or commercial monetization.
Target Markets
- Software engineers using Claude Code
- Teams operating multi-agent or skill-based AI coding workflows
- Developers building software factories and wanting observability into agent orchestration
- Maintainers debugging token usage, latency, and context behavior in agent harnesses
- Diagnosing unexpected behavior in non-trivial Claude Code setups with skills, subagents, and memory files.
- Finding latency, token, and context inefficiencies in AI-assisted software-development workflows.
- Understanding orchestration and validation call trees visually after a coding-agent session.
- Writing automated expectations or checks against plain-JSONL traces from a software factory.