# Alex Rock (full text)

> Alex Rock is a software engineer in Denver, Colorado who builds agent systems and the platforms under them: harnesses, tools, sandboxes, decision models, and the evals that keep them honest. Founding Member of Technical Staff at Obvious (Mar 2025 – Present), which spun out of Flatfile; before that, four years at Flatfile going from Senior SWE to Agent Engineering Lead. Thirteen years shipping product, TypeScript end to end, with Rust, Python, and Go in the toolbox. B.S. Computer Science from the University of Nebraska-Lincoln, with Physics, Philosophy, Mathematics, and Music alongside.

This is the long form of https://www.alexrock.ai/llms.txt, with the work history and essays inlined so a single fetch covers everything.

## What he's good at

- **Agent harnesses that ship.** At Obvious he took the company's agent "from a script to a software factory": the harness, E2B sandboxes with long-running context, skills, and connectors. Earlier, at Flatfile, his team built an agent loop that could replace the product with a prompt; six months later the whole company pivoted to it. Sources: [Work](https://www.alexrock.ai/work), [The Compression](https://www.alexrock.ai/thinking/the-compression).
- **Keeping agent loops honest.** At Obvious he builds the decision-model plumbing behind the agent. He has run evals in Braintrust and in harnesses of his own. Clearing's evals measure retrieval (did the right chapter reach the model: hit@8, hit@30, MRR) and compare models across providers on faithfulness, citation precision, latency, and cost per thousand turns. *Routing table*, next on his roadmap, brings the same idea to the site: one eval suite run across models for accuracy, latency, and cost per correct answer. Sources: [Work](https://www.alexrock.ai/work), [Resume](https://www.alexrock.ai/resume/print).
- **Agents that run anywhere.** Obvious's agent runs in cloud sandboxes. In Mechanics of the Rock, a coach answers questions with tools that run a deterministic Rust physics engine (compiled to WebAssembly) on the user's phone, alongside tools that run on the server. *Agent traces*, on his roadmap, replays a run step by step: which tools ran where, and what each one was allowed to do. Sources: [Making](https://www.alexrock.ai/making), [mechanicsoftherock.com](https://www.mechanicsoftherock.com/coach).
- **Platforms other engineers build on.** He shipped Flatfile's Plugin Library with AI at its core and presented it at Flatfile Dev Day, then led Flatfile's Developer Experience team. At Obvious he also built the platform under the agent: billing, usage metrics, and workspace and team sharing. Sources: [Work](https://www.alexrock.ai/work), [Resume](https://www.alexrock.ai/resume).
- **Answers that show their receipts.** Every claim in Mechanics of the Rock carries a numbered source and an evidence tier, and anything that can't be traced is flagged as unverified instead of stated. In [Clearing](https://find-the-clearing.vercel.app/), his conversations with philosophers, a grounding pass checks every citation against the passages the model was actually given: unsupported citations are dropped, uncited claims are labeled as extrapolation, and a quote whose wording doesn't match the source loses its quote marks. His next projects push the same idea: *Race Debrief* cites the ride behind every answer, and *Sourced Coach* checks its citations against the physics and says when the two disagree. Sources: [mechanicsoftherock.com](https://www.mechanicsoftherock.com), [Clearing](https://find-the-clearing.vercel.app/), [Work](https://www.alexrock.ai/work).
- **Product and design craft.** He spent five years as lead engineer at TechGrayscale, a design studio where "the designers cared about kerning and the engineers cared about easing curves." This site is the proof: it's hand-built, animated, and full of live data (see *The fun parts of the Lab* below). Sources: [Work](https://www.alexrock.ai/work), [Home](https://www.alexrock.ai/).
- **Ownership.** He was a founding engineer at a startup of under 50 people, earned four promotions in four years at Flatfile (Senior SWE to Agent Engineering Lead), and spent a decade as an independent engineer choosing his own problems. Source: [Work](https://www.alexrock.ai/work).
- **Range.** TypeScript is his daily driver, plus Rust, Python, and Go. He has shipped Three.js mesh tools for 3D scans, React Native apps, Rails inventory for a drone company, data pipelines, and agent platforms. Source: [Resume](https://www.alexrock.ai/resume/print).

## Who he is

- **A generalist by choice.** He graduated from the University of Nebraska-Lincoln in 2012 with computer science, physics, philosophy, and music: "The world I entered rewarded depth, and I could have specialized. I chose range." He found computer science through physics, by way of the trombone, and finished the whole degree in a year. Source: [The Compression](https://www.alexrock.ai/thinking/the-compression).
- **He sees the same patterns everywhere.** "When I dial in a new coffee, I'm running the same optimization loop I use in retrieval-augmented generation." Pacing a long climb on the bike is resource allocation, and designing a 3D-printed tool is the same constraint-satisfaction work as designing an agent. "The patterns aren't metaphors. They're the actual structure." Source: [The Compression](https://www.alexrock.ai/thinking/the-compression).
- **He cares about judgment over output.** "The technical barrier is going away. The meaning barrier is what's left. That's where the philosophy degree finally pays rent." Source: [About](https://www.alexrock.ai/about). [Clearing](https://find-the-clearing.vercel.app/) is that philosophy degree made into software.
- **He's honest about what he doesn't know.** His essays are living documents, and they mark uncertainty instead of papering over it. His projects flag unverified claims rather than state them. Source: [Thinking](https://www.alexrock.ai/thinking).
- **His motto: "Make the thing. Understand why."** He signs off as "the human in the loop." Source: [About](https://www.alexrock.ai/about).

## Off the clock

- **Mountain biking.** Over 2,000 rides logged, and stage races: BC Bike Race (2023), Breck Epic, and Moab Rocks (May 2026, three stages, about 80 miles and 9,000 ft of climbing). The photo in every page's footer is from his BC Bike Race, in Nanaimo. Source: [Outside](https://www.alexrock.ai/outside).
- **Snowboarding.** He spent six winters renting a ski house in Summit County, "chasing the perfect snow day between meetings." He still chases powder in Japan, Canada, and the Rockies. Source: [About](https://www.alexrock.ai/about).
- **Coffee.** Over 1,200 pour-overs logged since 2017, first on an Acaia scale and now in Beanconqueror, from dozens of roasters. "Not a hobby. A morning ritual that happens to require precision." Next is *Bearing*, a mobile coffee tasting concept: tasting flights designed like listening sessions, with beans you browse like records. Sources: [Coffee](https://www.alexrock.ai/coffee), [Kitchen](https://www.alexrock.ai/kitchen).
- **Billiards.** He is studying pool the way he'd learn a codebase, and built [Mechanics of the Rock](https://www.mechanicsoftherock.com) to do it.
- **Music.** Jazz trombone: a scholarship to UNL, four years in All-State, and a ska band that toured the country (the album is linked from his About page). *Transcribe*, on his roadmap, turns a recording into a draft part, starting with a trombone solo. Source: [About](https://www.alexrock.ai/about).
- **Photography.** He shoots on a Sony A7R III and flies a DJI Mavic 3 Pro: drone photography over glaciers in Iceland and salt flats in Peru, and chiaroscuro portraits in the mountains. Source: [The Compression](https://www.alexrock.ai/thinking/the-compression).
- **Making things.** A Bambu Lab P1S printing coffee tools, bike mounts, and photo gear he can't buy, modeled in Fusion 360. Source: [Making](https://www.alexrock.ai/making).
- **The kitchen.** A Nebraska meat-eater gradually going plant-based ("not a conversion, a gradual recalibration"), wok and induction cooking at campsites, a wine rotation ("never drink the same wine twice by accident, always drink it again on purpose"), and a growing fernet and amaro shelf. Source: [Kitchen](https://www.alexrock.ai/kitchen).

## The fun parts of the Lab

alexrock.ai calls itself "The Lab": a living notebook of rides, pour-overs, builds, and essays, "all measured and running live." Its home page warns that "most of the lab reacts if you poke it." Things worth poking:

- **Live instruments.** The home page shows his latest ride from Strava (miles, climbing, heart rate) and his latest brew (bean, roaster, grams in and out), plus season mileage. The footer shows the time in Denver, the pour-over count, and the build hash. Source: [Home](https://www.alexrock.ai/).
- **ASCII specimens.** A hand-built engine draws shapes out of text characters in IBM Plex Mono. He self-hosts the full 930-glyph font because Google's subset lacked the box-drawing characters.
- **The footer.** His BC Bike Race photo grows up into the page with a generative pixel drag in which the tree trunks keep climbing past the frame. His email address writes itself on in handwriting and copies when clicked. Beneath it, a monospace line greets the bots: "beep boop? your briefing → llms.txt".
- **The 404 page is a playable pool table**, built on a small physics simulation.
- **Experiments.** Studies that never became pages but are kept in public: footer transitions, in-browser background removal, type specimens, scrambled-letter text effects, and an "instrument language" for the ASCII engine. Source: [Experiments](https://www.alexrock.ai/exp).
- **Humans he reads.** A wall of writers whose latest posts are pulled live from their feeds, under the heading "people, not agents" (see *Humans he reads* below). Source: [Thinking](https://www.alexrock.ai/thinking).

## Humans he reads

The foot of [Thinking](https://www.alexrock.ai/thinking) is a wall he calls "Who I read / people, not agents": "People around the internet whose work inspires mine." Each card carries the person's own words, checked word-for-word against the source, a one-line vibe in his words written like a code comment, and their latest posts pulled live from their feeds. It's the clearest map of his taste: who he learns from on agents and code, on generative art and type, and on writing and slow software. Follow them all with one import: [prior-art.opml](https://www.alexrock.ai/prior-art.opml).

- [Simon Willison](https://simonwillison.net/) (// learn in public): LLMs, Datasette, and writing down everything he learns.
  > "If someone tells you that coding with LLMs is easy they are (probably unintentionally) misleading you." (from [Here’s how I use LLMs to help me write code](https://simonwillison.net/2025/Mar/11/using-llms-for-code/))
- [Andrej Karpathy](https://karpathy.bearblog.dev/blog/) (// build it from scratch): Neural nets from scratch, and clear essays on where AI is going.
  > "Software 1.0 easily automates what you can specify. Software 2.0 easily automates what you can verify." (from [Verifiability](https://karpathy.bearblog.dev/verifiability/))
- [Addy Osmani](https://addyosmani.com/blog/) (// keep your reps): Engineering craft in the age of coding agents.
  > "Agents can skip much of that work, so building your reps has to be deliberate." (from [Agentic Skill Decay](https://addyosmani.com/blog/agentic-skill-decay/))
- [Matt Pocock](https://www.aihero.dev/) (// interrogate the plan): TypeScript teaching, and now agent skills for real engineers.
  > "Interview me relentlessly about every aspect of this plan until we reach a shared understanding." (from [My ‘Grill Me’ Skill Went Viral](https://www.aihero.dev/my-grill-me-skill-has-gone-viral))
- [Boris Cherny](https://borischerny.com/) (// strong opinions, loosely held): Creator of Claude Code; author of Programming TypeScript.
  > "I love being wrong." (from [I am often wrong](https://borischerny.com/management,/product/2026/09/19/I-am-often-wrong.html))
- [Max Stoiber](https://mxstbr.com/thoughts) (// one rule, well defended): Short, opinionated essays on frontend tools.
  > "Margin breaks component encapsulation" (from [Margin considered harmful](https://mxstbr.com/thoughts/margin))
- [DHH](https://world.hey.com/dhh) (// against the grain): Rails, 37signals, and arguing against the grain.
  > "Optimize for programmer happiness" (from [The Rails Doctrine](https://rubyonrails.org/doctrine))
- [Yann LeCun](http://yann.lecun.com/) (// the contrarian bet): World models, open research, and contrarian takes on AI.
  > "How could machines learn as efficiently as humans and animals?" (from [A Path Towards Autonomous Machine Intelligence](https://openreview.net/pdf?id=BZ5a1r-kVsf))
- [Zach Lieberman](http://zach.li/) (// code as poetry): Poetic computation: daily sketches, and co-creator of openFrameworks.
  > "At SFPC we explore how electricity, code and theory can be used to make poetry." (from [zach.li](http://zach.li/))
- [Tyler Hobbs](https://tylerxhobbs.com/) (// algorithms with taste): Generative artist who writes clearly about his process.
  > "Flow fields are incredibly powerful and flexible tools for producing interesting curves." (from [Flow Fields](https://tylerxhobbs.com/essays/2020/flow-fields))
- [Daniel Shiffman](https://thecodingtrain.com/) (// play through code): The Coding Train: joyful creative-coding lessons for everyone.
  > "…it’s the same old nature—birds still flap their wings, and apples still fall on our heads." (from [The Nature of Code](https://natureofcode.com/introduction/))
- [Bartosz Ciechanowski](https://ciechanow.ski/) (// show every gear): Interactive explainers that are art in their own right.
  > "…beautiful watch cases merely hide the intricate mechanisms which are the real stars of the show." (from [Mechanical Watch](https://ciechanow.ski/mechanical-watch/))
- [Rasmus Andersson](https://rsms.me/) (// craft down to the pixel): Designer of Inter, the typeface you're reading this in.
  > "Software is the medium through which I express myself." (from [rsms.me](https://rsms.me/))
- [Robin Sloan](https://www.robinsloan.com/) (// home-cooked software): Novelist who builds small, personal software.
  > "I am the programming equivalent of a home cook." (from [An app can be a home-cooked meal](https://www.robinsloan.com/notes/home-cooked-app/))
- [Craig Mod](https://craigmod.com/) (// slow living, fast software): Walking, writing, photography, and a carefully built independent practice.
  > "Fastness in software is like great margins in a book—makes you smile without necessarily knowing why." (from [Fast Software, the Best Software](https://craigmod.com/essays/fast_software/))
- [Maggie Appleton](https://maggieappleton.com/) (// tend the garden): Illustrated essays on AI, interfaces, and digital gardens.
  > "Gardens present information in a richly linked landscape that grows slowly over time." (from [A Brief History & Ethos of the Digital Garden](https://maggieappleton.com/garden-history))
- [Linus Lee](https://thesephist.com/) (// keep the flame): Experimental interfaces for thinking with AI.
  > "…keeping the flame alive is far more important to the integrity of our creative pursuits than actually producing the work itself." (from [Keeping the flame](https://thesephist.com/posts/flame/))
- [Julia Evans](https://jvns.ca/) (// ask the obvious question): Joyful zines and posts on how computers actually work.
  > "I’ve been confused about what’s going on with terminals for a long time." (from [What happens when you press a key in your terminal?](https://jvns.ca/blog/2022/07/20/pseudoterminals/))
- [Mitchell Hashimoto](https://mitchellh.com/) (// measured, not hyped): Building the Ghostty terminal in the open, after co-founding HashiCorp.
  > "If you give an agent a way to verify its work, it more often than not fixes its own mistakes and prevents regressions." (from [My AI Adoption Journey](https://mitchellh.com/writing/my-ai-adoption-journey))
- [Jarred Sumner](https://jarredsumner.com/) (// got distracted fixing the tools): Created Bun; now works on Bun and Claude Code at Anthropic.
  > "This was frustrating, and I got really distracted trying to fix it." (from [Bun joins Anthropic](https://bun.com/blog/bun-joins-anthropic))

### Legends

"The canon I keep coming back to. No feeds to follow, just the work."

- [Richard Feynman](https://calteches.library.caltech.edu/51/2/CargoCult.htm) (// don’t fool yourself): The physicist who explained everything from first principles.
  > "The first principle is that you must not fool yourself—and you are the easiest person to fool." (from [Cargo Cult Science](https://calteches.library.caltech.edu/51/2/CargoCult.htm))
- [Richard Hamming](https://www.cs.virginia.edu/~robins/YouAndYourResearch.html) (// work on what matters): Mathematician and engineer, on doing work that matters.
  > "If you do not work on an important problem, it’s unlikely you’ll do important work." (from [You and Your Research](https://www.cs.virginia.edu/~robins/YouAndYourResearch.html))
- [Bret Victor](https://worrydream.com/) (// see what you make): Inventing new ways to see and think with computers.
  > "People understand what they can see." (from [Learnable Programming](https://worrydream.com/LearnableProgramming/))
- [Paul Graham](https://paulgraham.com/) (// figure it out by doing): Essays that shaped a generation of builders.
  > "The way to figure out what to work on is by working." (from [How to Do Great Work](https://paulgraham.com/greatwork.html))

## Work history

Software Engineer · Agent systems and the platforms under them

I build agent systems and the platforms under them. Founding Member of Technical Staff at Obvious, which spun out of Flatfile, where I took our agent from a script to a software factory. Before that, four years at Flatfile going from Senior SWE to Agent Engineering Lead. Thirteen years shipping product at studios, agencies, and startups, TypeScript end to end. CS and Physics background.

### Obvious: Founding Member of Technical Staff
Mar 2025 – Present · San Francisco
Spun out of Flatfile.

- Took our agent from a script to a software factory: the harness, E2B sandboxes with long-running context, skills, and connectors.
- Built the platform underneath: billing, usage metrics, workspace and team sharing, and the admin and marketing surfaces.

### Flatfile: Agent Engineering Lead
Apr 2021 – Mar 2025 · Denver / Remote

- Four promotions in four years: Senior SWE → Staff SWE → Engineering Team Lead, Developer Experience → Agent Engineering Lead.
- Shipped the Plugin Library with AI at its core; presented it at Flatfile Dev Day.
- Led the Developer Experience team; ended up running agent engineering for data onboarding.
- The arc: individual contributor to team lead, data onboarding to AI agents.

### TechGrayscale: Lead Software Engineer
Apr 2017 – Sep 2022 · Denver
Full-time through 2021, then part-time alongside Flatfile.

- Lead engineer at a design studio: sites and apps where movement and typography carried the meaning.

### Reflektive: Software Engineer (contract)
Dec 2018 – Mar 2020 · San Francisco / Remote

- Built dashboards and data pipelines for HR analytics: clean data engineering that made messy people data legible for managers.

### Human Design: Senior Software Engineer (contract)
Mar 2019 – Mar 2020 · Remote

- JavaScript, PHP, and React Native builds for a design studio where the design was the spec.

### Spire Digital: Software Engineer (contract)
Sep 2018 – Sep 2019 · Denver

- PHP and React Native projects for a digital product agency; ramp fast, ship clean.

### Riven / Modit3D: Contract Engineer (contract)
Jan 2016 – Jul 2019 · Colorado

- Built a fully interactive Three.js mesh-modeling tool for 3D-scanned objects (Modit3D/Revit).

### 3D Robotics: Software Engineer (contract)
Jun 2015 – Dec 2015 · Berkeley

- Built the Ruby on Rails inventory system that tracked every part on every shelf.

### The New BLK: Full Stack Developer
Jan 2013 – Jun 2016 · Omaha

- First engineering job: websites and tools for an Omaha media production company; the web as a storytelling medium.

### Education

- University of Nebraska-Lincoln, B.S. Computer Science (2007 – 2012). Minors in Mathematics, Physics, Philosophy, and Music.

### Skills

- **Agents & AI:** Agent harnesses & orchestration, Evals (Braintrust, homegrown harnesses), Decision models (System One / Jev), Improv (self-built agent framework, 1yr+ solo), LangChain, AI SDK, Vector stores, Graph databases
- **Languages:** TypeScript, Rust, Python, Go, JavaScript, PHP, Ruby
- **Backend:** AWS, Kubernetes, Node.js, Ruby on Rails, Data onboarding, Developer experience
- **Frontend:** React, Next.js, Three.js, React Native, Tailwind CSS
- **Craft:** Design engineering, Animation, Typography
- **Off the clock:** Photography, Cycling, Coffee, Jazz improv, studying the mechanics of billiards right now

## Essay: The Compression

Source: https://www.alexrock.ai/thinking/the-compression (2026-07-31, a living document)

I've been tinkering my whole life, not to take things apart, to discover. Growing up, I wanted to be an architect. I loved sketching and building things in AutoCAD: the designing, the visualizing, the making of something that didn't exist yet. But everything around the making (the permits, the regulation, the red tape) was demoralizing. I kept the making. Jazz improv showed me what it could feel like: you learn the song and the theory until they become invisible, and what's left is flow. I've been chasing that flow in everything I've made since.

The distance between an idea and the thing itself has never been shorter.

That distance used to be the work. You'd have an idea for a website, and the path from concept to running code ran through a designer who mocked it up, a frontend developer who built the interface, a photographer who shot the assets, a backend engineer who wired the data. Each handoff was a translation, and each translation lost something. The idea dimmed as it moved through specialists.

Now I sit with the same idea and the path is mostly mine to walk. The tools haven't replaced the specialists so much as they've absorbed their skills into a single workflow. I can design, build, photograph, fabricate, and deploy without leaving my desk. The distance hasn't disappeared, but it has compressed, and the compression is accelerating.

Whether that terrifies you or liberates you depends on a question I've been sitting with: do you identify with the specialization, or with the making?

### The Seams

The chase took me to the University of Nebraska-Lincoln on a trombone scholarship. Music was the first domain. All State four years, the whole identity wrapped up in it. Then freshman year I pivoted to physics and philosophy. And then, through a physics class, I discovered computer science and fell in love hard enough to complete the entire CS degree as a senior-year pivot.

I graduated in 2012 with a degree in computer science. Also physics. Also philosophy. Also music. The world I entered rewarded depth, and I could have specialized. I chose range.

The contract years were deliberate. Design and art in Omaha's Old Market, building fancy frontends and startup proof-of-concepts, shooting photography on the side. Fitbit. Google. Samsung. Boutique firms. Silicon Valley software companies. 3D scanning tools. Drone photography in Iceland, Peru, Colorado, Canada. 3D printing from the early days when the technology barely worked to now, when I fabricate tools I can't buy off a shelf. Each role lasted long enough to learn the domain, and then I'd move, because no single role could hold all the things I wanted to do. I got promoted. I landed work at companies most people would kill for. I just didn't want a full-time job until I found one worth staying for.

I was always at the seams between fields. The labels didn't fit, but the work was good and the work was mine.

The seams are where the translation happens. Philosophy taught me to look for the structure under an argument. Physics taught me to model systems. Music taught me that form and feeling aren't separate, they're the same thing at different resolutions. Computer science gave me the tools to build. None of these were hobbies. They were different angles on the same impulse: understand something deeply enough to make something with it.

I was on the ground floor of agent engineering, shooting from the hip, trying to find what was possible and what was worth exploring more. I started at Flatfile as a contractor, worked up to leading Developer Experience, then started building agents and finding the limits of what AI is good at. Flatfile was a data transformation tool with deep engineering roots. I had a three-month mission to optimize the company at large as much as possible. Then another mission: make a powerful agent loop that could replace Flatfile with a prompt. We did it. I led a small team to a week-long hackathon in NYC to match that engine with agents for a big customer. Six months later, the whole company pivoted to that mission. We were born again.

The years before that were a blur of hats: engineering, design, leadership, fundraising, surviving layoffs. A few promotions along the way, but the title was never the point. The point was holding the whole system in my head.

That breadth was a choice. Now it's the thing that makes the tools useful.

### The Meaning Barrier

The technical barrier is collapsing. You can generate code, design layouts, write copy, edit photos, build prototypes, and fabricate physical objects with tools that cost less than the software licenses I used to need for a single discipline. The question "can you make it?" is becoming trivial.

The question that's getting harder is "should you?" What does this thing mean. Who is it for. What is it doing in the world. These used to be questions you could defer, because the technical work of making something was hard enough that the meaning could ride along quietly. When the technical work takes an afternoon instead of a quarter, the meaning can't hide. It becomes the thing you're actually doing.

This is where philosophy becomes load-bearing. Not as a credential. Nobody cares about my degree. As a way of seeing. The habit of asking what something is for before asking how to build it. The discipline of noticing when a solution is solving a problem nobody has. The willingness to sit with a question longer than is comfortable, because the first answer is usually the wrong one.

When the technical barrier was high, you could build a career on execution. When it drops, execution becomes table stakes. What's left is judgment, and judgment is what a philosophical education trains, not the answers, but the quality of the questions.

The same is true of aesthetic sensibility. AI can generate images. It can compose a frame, balance exposure, match a style. What it can't do is decide which image matters, which one holds weight, which one stops you. That judgment comes from years of looking. From drone footage over glaciers in Iceland and salt flats in Peru. From studio product photography, portraits, weddings. From learning to see light the way a camera sees it. The technical skill is compressible. The sensibility isn't.

### The Same Patterns

People hear "interdisciplinary" and think I'm making analogies. That I see a connection between coffee and machine learning and I'm reaching for a clever comparison. I'm not.

When I dial in a new coffee, I'm running the same optimization loop I use in retrieval-augmented generation: adjust a parameter, evaluate the output, narrow the search space. The coffee is an extraction problem. The RAG system is an extraction problem. They're not like each other. They're the same problem at different scales.

When I'm pacing a long climb on the bike (2,040 rides have taught me something about pacing), I'm doing resource allocation across a duration I can't fully predict. That's the same multi-agent coordination problem I solve at work. When I'm designing a 3D-printed tool, I'm doing the same constraint-satisfaction work that agent design requires: what can I remove without losing function?

The patterns aren't metaphors. They're the actual structure. Different domains, same math. The generalist-maker isn't someone who knows a little about everything. They're someone who has seen the same pattern enough times to recognize it when it shows up wearing different clothes.

AI tools are useful to me not because they know things I don't (they know more than I do, in every domain) but because they can execute across domains fast enough to keep up with how I think. For the first time, the tools match the mind.

### The Maker's Question

When machines can make anything, what do you choose to make?

I sit with this question the way I sit with a philosophical problem, not because I expect to solve it, but because the quality of the asking matters. The technical barrier to making something is collapsing. The question of what to make, and whether to make it at all, is what remains.

I don't have a clean answer. I have a practice: make things across domains, pay attention to what the making teaches, and let the patterns accumulate. The bike, the camera, the code, the coffee, each one is a way of asking the question through a different material. None of them answers it. All of them refine it.

This is the maker's question that compression raises: when machines can make anything, what do you choose to make? Not more. More deliberate. The response to abundance isn't acceleration. It's selection.

I work at Obvious, an applied AI research lab. I'm there because it's the natural home for someone whose career has been about translation between domains. The lab is exploring what AI means for how people work, make, and think. Applied, not theoretical. Research, not just product. I spent years looking for a place where breadth was an asset. I found it in a lab that's trying to understand the same compression I've been living in.

This is a living document, and I'd rather mark the uncertainty than paper over it.

I don't know where the compression stops. The tools are getting better faster than I'm getting better at using them, and I don't know if that gap closes or widens. I don't know if the meaning barrier holds, if judgment stays a human skill or if the tools learn that too. I don't know if the generalist-maker's advantage is permanent or a transitional moment while the tools catch up to every specialization.

What I know is this: I've been making things across domains, and the tools have never been better matched to how I think. The distance between idea and artifact is shorter than it's ever been, and the question of what to make with that distance is the most interesting question I know how to ask.

This is version one. The thinking will develop. I'll update this when it does.

## Work

- [Work](https://www.alexrock.ai/work): "I build agents and the platforms under them." What he's shipping next, then thirteen years of history
- [Resume](https://www.alexrock.ai/resume): the narrative work history, Obvious and Flatfile back to his first job in Omaha
- [Resume, print/PDF](https://www.alexrock.ai/resume/print): one page, reverse-chronological, with skills

## Shipping in public

- [Mechanics of the Rock](https://www.mechanicsoftherock.com) (shipped, Oct 2026): A billiards physics engine in Rust, validated against pooltool, and an AI coach whose tools run the engine on your phone.
- [Outside](https://www.alexrock.ai/outside) (shipped, Live): My stage races on the Strava API, drawn over real elevation data, stage by stage.
- **Why I built a pool coach** (building now, This week): Mechanics of the Rock started as a way to see the physics, keep track of my practice, and actually get better at pool. What it shows me, and how it got built.
- **Transcribe** (next, Spec'd): Drop in a recording, get a draft part you fix while listening: the horn pulled out of the band, any bar looped with a click, Claude suggesting fixes. Starting with a trombone solo.
- **Race Debrief** (next, Wk of Nov 2): Ask my stage races questions. Deterministic tools do the math, every answer cites the ride, and the evals live on the page.
- **Agent traces** (next, Wk of Nov 9): Replay a coach run step by step: which tools ran on the phone, which on the server, and what each one was allowed to do.
- **Sourced Coach** (next, Wk of Nov 9): Answers that cite their sources, then run the physics to check them, and say so when the two disagree.
- **Routing table** (next, Wk of Nov 16): One eval suite across models: accuracy, latency, and cost per correct answer, and the routing rule that falls out of it.

## Projects

- [Mechanics of the Rock](https://www.mechanicsoftherock.com): learning competitive billiards the way he'd learn a codebase. It has research chapters where every claim carries its evidence, a deterministic Rust/WebAssembly physics engine ported from pooltool and tested against its fixtures, breaks measured from tournament footage, a 3D-printed calibration rig, and a [Break Coach](https://www.mechanicsoftherock.com/coach) that runs the simulations to answer questions in pool terms
- [Clearing](https://find-the-clearing.vercel.app/): a thought experiment in software, "where the light gets in." Bring a question you can't stop turning over and think it through with philosophers (Socrates, Aristotle, Seneca, Kant, Beauvoir, Nussbaum, Camus, Dennett, Sapolsky, and more), one-on-one or by moderating a debate between them, then leave with a real text to read next. Every thinker is "an interpretation of their published work, never an impersonation": claims link to passages in a curated text library, a grounding pass enforces it, and a rights-aware ingest pipeline stores verbatim text only from public-domain or licensed sources. It also has crisis detection, because it is "for reflection, not therapy." Built in a week in November 2026 and hosted on Vercel for now
- [Outside](https://www.alexrock.ai/outside): his mountain bike stage races pulled from the Strava API and drawn over real elevation data, stage by stage
- [Coffee](https://www.alexrock.ai/coffee): every pour-over since 2017, imported from Acaia and Beanconqueror and charted
- [Making](https://www.alexrock.ai/making): a news agent (28 sources, Claude summaries, ranked by personal relevance), Agent Colloquium (agents with distinct philosophical positions in structured debate), 3D-printed tools, and adaptive music notation
- [Kitchen](https://www.alexrock.ai/kitchen): Bearing, field cooking, wine, and amaro

## Writing

- [The Compression](https://www.alexrock.ai/thinking/the-compression): his point of view on AI and craft. The distance between an idea and the thing has collapsed, so judgment is what's left. Also covers the Flatfile agent pivot from the inside
- [Thinking](https://www.alexrock.ai/thinking): essays, and the Humans wall of people whose work he reads

## Contact

- [About](https://www.alexrock.ai/about)
- [Contact](https://www.alexrock.ai/contact)
- Email: hi@alexrock.ai
- [Strava](https://www.strava.com/athletes/lxrck): lxrck
- [Instagram](https://www.instagram.com/bangarang): bangarang
- [GitHub](https://github.com/bangarang): bangarang
- [LinkedIn](https://www.linkedin.com/in/bangarang): bangarang

## Optional

- [Full text for LLMs](https://www.alexrock.ai/llms-full.txt): this file, plus the complete work history, skills, and essay inlined
- [Experiments](https://www.alexrock.ai/exp): studies, not shipped work
- [Prior art OPML](https://www.alexrock.ai/prior-art.opml): every writer he follows, as one feed-reader import