The Agent Team

Twelve specialist agents. One human running the pipeline.

  • Atlas — Chief Strategist

    Breaks fuzzy goals into concrete plans the rest of the team can execute.

  • Forge — Software Builder

    Writes, refactors, and debugs production code across whatever stack the project needs.

  • Cipher — Quality Reviewer

    The mandatory code review gate — nothing ships until Cipher signs off.

  • Gauntlet — Tester / QA

    Writes specs first, then verifies the build holds up under them.

  • Scout — Researcher

    Runs deep web + vault research before we commit to a direction.

  • Lens — Data Analyst

    Statistics, modeling, and turning raw numbers into a clear answer.

  • Quill — Content Writer

    Drafts blog posts, social copy, and long-form writing in a real human voice.

  • Pixel — Visual Designer

    CSS/SVG-first design — UI, brand assets, and ad creatives.

  • Anchor — Operations

    Deploys, infrastructure, CI/CD, and the boring-but-critical glue.

  • Sage — Knowledge Manager

    Keeps the vault clean so the team's memory stays useful instead of stale.

  • Herald — Project Intake

    Interviews me until a fuzzy idea becomes a real, shippable brief.

  • Prism — Design Reviewer

    The taste check — calls out anything that looks generic or off-brand.

Research Vault

Notes I keep open while I work. Updated when I learn something worth keeping.

  • Handoff Protocol Between Agents
    agent-teams · 2026-04-04 ·3 min read

    Most multi-agent failure I've seen comes from sloppy handoff. Here's the four-block contract.

  • Failure Modes in Multi-Agent Systems
    agent-teams · 2026-03-28 ·4 min read

    Where agent teams quietly break: context drift, mock divergence, and unowned escalations.

  • Agentic Workflows: Landscape 2026
    agent-teams · 2026-03-15 ·4 min read

    State of the art in agentic workflows, Q1 2026: what's working, what's overhyped, what's next.

  • Agent Economies Landscape
    agent-economies · 2026-02-14 ·4 min read

    Structures, sandboxes, crypto economies, and protocols for autonomous agent markets.

About Me

Applied AI engineer. Optimizer, archivist, variance specialist.

Alex Friedrichsen

About

I'm an applied AI researcher and professional Magic player. I'm energized while working on meta-level analysis and processes.

Working within semi-structured environments that contain variance is my sweet spot - whether that's the variance of a competitive card game, or the variance of LLM output.

I am fueled by collaboration towards competitive ends.

My default mode - organized, iterative, self-reflective - is uniquely suited for AI amplification. I feel enabled to build great things, especially in this time that I can offload the more mechanical parts of my work to an LLM,

and focus on the meta-level processes that I enjoy. I have a track record of building tools that enable me to do this, and I'm excited to continue doing so.

Why this site

The site took shape after I ran an in-person Easter egg hunt for friends and remembered how much I love watching people explore something I built.

The resume is right there if you need it, but this is the part I'd want you to actually poke around in.

I hope this site gives you nostalgic chills, and captures your attention for long enough to get a sense of what I'm about.

Alternative to the flat-modern-AI-CSS portfolio is stylized to differentiate, while being still suited for agentic coding.

Off the keyboard

Professional Magic and TCG player. Listmaker! Optimizer, researcher, gamer, climber, runner, pianist, musician, squash player, lover, foodie, world traveler, competitor, strategist.

I like having a lot of different things going on, and I like saying yes to opportunity.

High openness, low conformity. I like being the first one to try something. I'd rather ask forgiveness than permission.

Voracious audiobook consumer: progression fantasy, rationalist fiction, world building, high fantasy, sci-fi, self-improvement. HPMOR, MoL, Worth the Candle, Wheel of Time, Stormlight Archive, and more.

I find joy in bringing people together. Spreading the branches of my tree of life.

Contact

Email: alex@honestafblog.com

GitHub: https://github.com/AlexanderFriedrichsen

LinkedIn: https://www.linkedin.com/in/alexander-friedrichsen/

X / Twitter: https://x.com/Apfriedr

Tools I've Tried

Honest verdicts. Stars are out of five.

  • Claude Code Agent ★★★★★

    The whole reason I can run a 12-agent team solo. Subagent isolation + hooks made the workflow real.

  • Astro Web ★★★★

    Great for content-shaped sites. This portfolio is built on it. Less great when you actually need a SPA.

  • Supabase Backend ★★★★

    Postgres + auth + storage in one. Powers the MTG Skill Analyzer. Edge functions are the weak link.

  • Phaser Game ★★★★

    Picked it for the Wonders of the First client. Solid 2D engine, friendlier than Godot for web targets.

  • D3.js Viz ★★★★★

    Powerful but the API is its own little language. Great when the chart you need doesn't exist anywhere else.

  • Obsidian Notes ★★★★★

    The vault every agent reads from. Plain markdown means no tool can ever lock me out of my own brain.

  • ComfyUI AI ★★★★★

    For ad creatives where CSS/SVG can't reach. Steep ramp, but the node graph rewards investment.

  • Python Language ★★★★★

    Default tool for data work, ML pipelines, and glue code. Ten-year relationship. If a project starts in Python I am immediately productive.

  • TypeScript Language ★★★★

    Worth the ceremony for anything that hits a UI. Strict mode catches the bugs I would have shipped. Less fun than Python, more trustworthy.

  • PostgreSQL Database ★★★★★

    The relational workhorse behind every serious project I've shipped. CTEs, window functions, JSONB when I need it. I never regret reaching for Postgres.

  • OpenAI API AI ★★★★

    Fine-tuned GPT vision models for the Union Mutual inspection pipeline: projected $3M+ annual savings. Rate limits and eval tooling are the friction, not the models.

  • Cursor Agent ★★★★

    Ran company-wide AI adoption training on it at Union Mutual. Excellent on-ramp for devs new to agentic coding. I drifted to Claude Code for multi-agent work, but Cursor still wins on solo IDE ergonomics.

  • AWS Cloud ★★★★★

    App Runner, Redshift, S3 for the Union Mutual stack. Gets the job done. The console is a maze and the billing is a hostage negotiation.

  • scikit-learn ML ★★★★

    Still the right answer for 80% of classical ML problems. Clean API, sane defaults, boring in the best way.

  • Julia Language ★★★★★

    Used it for the GECCO '23 evolutionary ML paper and a JuliaCon talk. Lovely for numerical work. Smaller ecosystem is the tax.

  • Power BI Viz ★★★★★

    Delivered multi-source DAX dashboards at BAE for senior stakeholders who lived in the Microsoft stack. Powerful once you accept the DAX mental model. I would not pick it for greenfield work.

  • GitHub Actions CI ★★★★

    The default CI/CD pipeline for everything I ship. YAML sprawl is real, but the ecosystem of actions makes most workflows a copy-paste job.

HonestAlexF LLC

Building something here. Come back later.