Aggregates local JSONL transcripts and cache snapshots (cached tokens included). Cursor is tracked server-side, so it exposes no local token counters.
Token pyramid · Jellyfish Q1 2026
10.68B
28x P90 · observed here
380M+ / mo
Top 10% · 90th percentile and above
51M to 380M / mo
Median to 90th percentile
Under 51M / mo
Below the median developer
Jellyfish publishes only the 50th (51M) and 90th (380M) percentile token figures, so the pyramid shows those three bands and does not invent finer tiers. The top block is this portfolio's 30-day total.
Jellyfish analyzed AI coding activity from 12,000 developers across 200 companies in Q1 2026. These are reference rates, not a global ranking.
Multi-agent orchestration, deterministic validation, and autonomous pipelines.
ARCHITECTURE DIAGRAMAFAW FLOW
Technical Director
Defines objectives, writes prompts, and decides what to integrate
Agent A
Dedicated branch & worktree
Agent B
Dedicated branch & worktree
Agent C
Dedicated branch & worktree
Deterministic Validation
Tests, CI, red-first, mutation testing, and state checks
Human Review & Approval
AFAW: Human-Directed Orchestration
A terminal per agent with targeted roles and tasks. Agents write code in isolated git worktrees; automated suites verify correctness and regression safety before merge.
Autonomous lifecycle with GitLab Duo, self-healing CI/CD, rule-based integration decisions and continuous deployment. Experiments ran for 48 hours, with successful autonomous merge requests executed by the agent at an approximate cost of 420 GitLab credits (USD 420).
# Agustin Diaz-Cano - Software Engineer
## AI Agent Usage (Last 30 Days)
- **Combined Total**: ~10.68 Billion tokens across 339 sessions (2026-10-09 to 2026-11-07).
- **Per-Agent Breakdown**: Claude Code (~7.91B tokens, 139 sessions) | Antigravity (~1689.0M tokens, 114 sessions) | Codex (~694.6M tokens, 61 sessions) | Meta Muse (~382.3M tokens, 23 sessions) | Grok (~5.4M tokens, 2 sessions) | Cursor (server-side only, 0 local transcript sessions, no local token counters).
- **Reference Comparisons**: The combined count is approximately 209 months (~17.5 years) at 51M tokens/month, or 28 months (~2.3 years) at 380M tokens/month. It is numerically comparable to one month at the median rate for ~209 developers.
- **Dashboard**: [AI Usage dashboard](https://www.agustindiazcano.com/agents-usage) with the per-agent breakdown, Jellyfish token pyramid and AFAW methodology.
- **Benchmark Source**: [Jellyfish Q1 2026 study](https://jellyfish.co/blog/is-tokenmaxxing-cost-effective-new-data-from-jellyfish-explains/) reported 51M tokens/month at the median and 380M at the 90th percentile among 12,000 developers at 200 companies. These are reference rates, not a global ranking; accounting methods and coverage may differ.
## About
Software Engineer | Python Developer with 4+ years of experience in software development, specializing in AI/ML and Backend Development (Python).
I'm currently an MS Candidate, Systems Engineering, at UTN, focused on Artificial Intelligence and Data Analysis. My technical focus is building backend architectures that integrate GenAI in production-realistic conditions, not just prototypes. I'm developing projects with Python, Retrieval-Augmented Generation (RAG), and the Model Context Protocol (MCP) to explore LLM orchestration and AI agent design.
Previously, Sole QA Owner and quality authority for Grupo GDU's e-commerce platforms. Responsible for the entire quality strategy of web and mobile across multiple high-traffic sites, covering a multi-million dollar online channel with hundreds of thousands of users. Held full Go/No-Go authority on production releases with no dedicated QA team. During my tenure, the platforms—previously unranked—achieved nominations for the eCommerce Awards Uruguay in 2024 and 2025.
Transitioned from zero coding background in the pre-AI era. Wrote my first line of code in November 2020, received a job offer on May 14, 2021, and shipped my first production PR in June 2021 at an international fintech. Participated in more than 10 projects that reached production.
In game development, I created full modifications for the Men of War series as a solo developer, reaching over 200,000 downloads. Only a small percentage of Steam games, mods, and mobile apps (typically under 2-5%, and often closer to 1% for the 100k threshold) ever reach that level of adoption.
## Full Context
**[Full Profile / System Context](/llms-full.txt)**
## Quick Links
- **Portfolio**: https://www.agustindiazcano.com
- **GitHub**: https://github.com/agustindiazcano
- **LinkedIn**: https://linkedin.com/in/agustindiazcano
- **Writing**: https://www.agustindiazcano.com/writing
Claude Code 
Antigravity 
Codex 
Meta Muse 
Grok 
Cursor