With a sufficiently advanced software factory, you can print & operate software at industrial scale. Distribution of all that software is a separate challenge (still figuring that part out). My nature, as an engineer, is to begin by optimizing the means of production.

The throughput of your factory depends on two things: the rate of input and token efficiency.

To improve the rate of input to my factory, I’ve focused on 3 areas:

1. Autonomy. I aim for my agents to be capable of working for hours or days off a single short prompt. Some iterative planning and specification is still required when I start large initiatives, but most of my steering happens at arms length.

2. Routines. I monitor a growing herd of agent routines. These require much less input from me: they iterate reliably toward some goal, evolving themselves over time.

3. Inspiration. I drink from a firehose of ideas so that I have plenty of high quality input for my factory. I’ve found X and HackerNews to be good watering holes, and I’m building a network of like-minded friends. I frequently feed my factory with the best open-source projects, blog posts, and research papers I encounter.

To improve token efficiency in my factory, I’ve focused on 2 areas:

1. Agent productivity. Your developer platform plays a pivotal role in the output of your factory. It’s well understood in software companies that strategic investments in shared foundations can increase velocity; they call it “developer productivity”. Agent productivity is the new developer productivity. Many old tools we used for optimizing human engineers are still relevant, but the field of agent productivity is brand new.

2. Agent coordination. Human teams pay a heavy coordination tax: scaling headcount doesn’t predictably increase output. Agent teams should in theory scale much more predictably, but it’s hard to get right. I have some experience here (I was #3 on the Tokscale leaderboard last week), and I believe shared memory is a core part of the solution.

Most pursuits boil down to modeling your domain, routing information, and iteratively building toward some fixed goal. If the means of production can be automated, put on your hard hat – it’s time to build a factory.

⌁ Ben

Well, I say – first of all, the return on creating art is immeasurable. ROI doesn’t matter here, software is just one of my creative outlets, and a monthly budget of $3k seems quite reasonable to me.

But, I continue – if you really want me to measure ROI: we must treat the token spend as capex, because it replaces human engineers. Software companies split engineering time into R&D (new products), COGS (operating existing products), and G&A (internal tools). Capitalized software becomes an intangible asset and amortizes over 3 years. EBITDA never sees that cost, so my ROI is unaffected by token spend. Q.E.D.

The funny look in their eye is beginning to fade, I think. I continue – another way to look at it is that I spend $3k/month on tokens. A fully loaded engineer is $25-40k/month. About a tenth of the price. Fifteen subscriptions, fifteen times the output. 150x software capex efficiency. Not bad. Plus, the token subsidies mean my $3k nets me $62k in API-equivalent value. Also, you must remember, high quality software is a durable store of value. Great software can even age like a fine wine, bottling the spirit of the times... my eyes glaze over with a faraway look.

I grab their arm. There’s a glint in my eye. I whisper maniacally – there’s a once in a lifetime discount happening at the OpenAI Intelligence Depot, and I brought 15 shopping carts. You have to go buy more shopping carts right now.

I begin to shout. Come with me! You must learn how to quickly load your shopping carts with tokens! You must race between the aisles, fill up your truck, and unload the tokens into your software factory! You must build your factory so that it consumes tokens efficiently, like a well-oiled machine! You must efficiently utilize all the tokens you have access to! Dream big! Dream more! More is more!

They begin to back away, slowly.

Harnessing tokens

When you set out to produce a lot of code, token efficiency is key. The best open-source models still lag closed models, and relative to subsidized costs, I believe closed models are still the better deal [0]. I prefer optimizing for quality over cost or speed, because I’ve found that when I use cheaper & dumber models, I end up spending more tokens (and human time) on cleanup.

When you operate a software factory, human time is the most constrained resource, so it’s critical to seek out forms of leverage on your time. The simplest source of leverage is to always use the ~best available model on the pareto frontier. After a bake-off between Fable and Sol, I decided to lean into Codex and use GPT 5.6 Sol (Max or Ultra) for every prompt in my factory [1].

I currently have 15 Codex Pro 20x subscriptions. I fully utilize all 15 subscriptions – I run out of my weekly usage in less than a day. If we estimate a 10x subsidy, that means I’m using $30k/month of premium inference productively, for only $3k/month.

Managing 15 subs is somewhat painful, so I built a simple CLI over the Codex App Server to track all my accounts. I’m working on a metaharness for Codex, Hra, which lets me unify all my subs and experiment with optimizations for my particular workflow.

[0] I track the pareto frontier of model capability on aicharts.io

[1] This is especially important when your agents are writing code. For non-coding tasks, I use cheaper models.

Harnessing writing

Early on, I decided to build a custom knowledge base (kb) for coding agents. kb autonomously percolates a knowledge graph, and gives agents tools to search efficiently: from simple fulltext search, to semantic search (using qmd and a local embedding model), to traversing the knowledge graph, to cross-referencing git history. Agents track plans and decisions in kb, which helps them coordinate and stay on track. I believe codebases have an intelligence of their own. The nature of code is that it encodes a belief system – and the best codebases also encode a clear picture of where they want to grow.

<writing style links, place model in smart codebase training distribution>

Harnessing autonomy

I set out to create a system that lets me do as little babysitting of my agents as possible. My agents run for hours at a time (sometimes days) from a single prompt. I check on them in a few focused sessions throughout the day: when I wake up, around lunchtime, and before bed.

First, I chose a simple tech stack that I know well: bun, NextJS, Vercel (domains, hosting, AI gateway, AI SDK), Convex, Turso, and PostHog. I established coding conventions, like enhanced type safety and property based testing. I built shared infrastructure for project boilerplate: UI, accounts, payments, observability, etc.

After I had a solid foundation for code quality and code reuse, I focused on the next highest leverage way to improve autonomy, which is to provide agents with tools to verify work themselves – this lets them cook for longer stretches without human intervention.

I decided to create a custom UI testing framework (direct), which lets agents simulate and verify app states & transitions quickly. direct sits at the seam between your app and the internet, and encodes an opinionated way to record UI flows and present them to agents efficiently.

Inspired by the success of direct in my factory, I decided to build a tool for programmatic browser automation (wrench). wrench is inspired by agent-browser’s derive-client skill – it lets agents grow an SDK for any target website. I use this primarily for personal automations, but it’s also useful for dev workflows where an official API path isn’t available.

Harnessing personal automation

After the foundation was in place, I decided to point my factory at personal infrastructure projects. I made the personal website of my dreams, with a living reading list that summarizes all the interesting links I encounter. The full contents of every article, video, or PDF get stored in my kb (so my agents have context on what I’m interested in), and published to my website, and it’s all automatic – I just send links to Grok Bot.

As the creator of Zo Computer – the first personal agent with a cloud computer – I’m deeply familiar with the space. I’ve long respected Benji Taylor and the Cursor team, so I was excited to try Bot. The computer use is slick, the agent is fast, and the interface is simple and delightful. It’s missing some things I appreciated about Zo, but after plugging in my homegrown tools, I’ve decided Bot is a worthy successor.

https://x.com/hraness/status/2090507003969917033 (TODO EMBED)

I’ve set up a handful of useful bots:

• News – I made an aggregator for the news I’m interested in at rough.day. News Bot sends me a summary each day.

• Wrench – In Bot’s browser, I signed into X, Substack, Bluesky, Instagram, Threads, Facebook, Reddit, YouTube, and Hacker News. I ask Wrench Bot to browse, summarize, and post to social media platforms, saving myself from brainrot.

• Read – I connected my kb with some guidance. Now I just send links to Read Bot, and it automatically updates my kb, summarizes the content for me to peruse, and publishes to my public reading list.

• Invest – I connected my Public account, asked Invest Bot to research and plan a portfolio strategy (saved to kb), and iterate 3 times a day. Fingers crossed.

• act60 – I’m building a suite of tools and resources for Puerto Rico Act 60. I generated a CLI, and hooked it up act60 bot, which now manages a lot of the administrivia for me.

• SEO – I hooked up PostHog, and now SEO Bot iteratively improves SEO across all my web properties.

• Stripe – I’m a big Stripe nerd, so I built stripehistory.com to catalog Stripe news since its inception. Stripe Bot notifies me when anything newsworthy happens.

• Atet – I’m building a tool for agentic media creation, atet.sh, named after the solar barque Ra sails on through the underworld. It iteratively creates images, diagrams, and videos, using a mix of generative and programmatic approaches. I chat with Atet to transmute raw assets and the seed of an idea into gold.

• People – I’m building an agentic rolodex, peopleblade.com. I ingested all my contacts (with useful data like how often I’ve interacted with a person) across iMessages, Apple Contacts, Gmail, LinkedIn, Telegram, WhatsApp, Instagram, and Facebook, and I’ve been progressively enriching thousands of contacts across my entire life. I chat with People Bot to manage my rolodex.

I left a cushy managerial job at Stripe in 2023, where I’d worked for over 8 years. I was fortunate to build Terminal as the first engineer, and it continues to be one of my proudest achievements.

When I started a company with , inference was all the rage, so we dreamed up a DAG-based platform. The idea was in the right ballpark, but only modestly successful after a year, so we shut it down and returned to the drawing board. We spent 6 months throwing spaghetti at the wall, then started building Zo. The manic excitement that emerges during that zero-to-one is truly the spice of life. I was drunk on Fable. Then I was drunk on Sol. I felt unstoppable.

During the depths of my AI psychosis, I had moments of mania: 3 all-nighters one week, on top of repeated mornings waking up too early with a head full of prompts. In the fog of war I began to see how the way I worked over the last 3 years was unsustainable. I’ve always suffered from an overactive brain. I always feel like the end is in sight, just around the corner, and if I grind just one more time I’ll see the light.

While I was reflecting on my burnout, an idea entered my head and I haven’t been able to ignore it. I realized I’ve been building tools to enable builders for 11 years. But in my heart of hearts, I’ve always been a musician, a creator, an artist, a bootstrapped indie hacker. I asked myself the hard question. When will I stop building the tools, and start building my own dream?

I can feel it in my bones that we’ve reached a bright new frontier with model capability. Just as I felt ahead of the curve last year, spending way too much money on Cursor, I have this edge today that most builders don’t have yet.

I’ve started building a software factory. Good software with a non-zero market size is a store of value. If you have a good process for producing good software cheaply and without much supervision, you should scale it up, produce a ton of good software, and figure out distribution so you can harvest the value.

One viable path to the one person $1B company is to create 100 $10M companies.

Three lessons for founders

As I close my chapter as Zo’s cofounder, I want to share three lessons, one for each of the past few years:

1. You must cultivate superhuman speed

The first big lesson, acquired through pain, is that there’s a real insanity that is required to win. Real personal sacrifices must be made. A mad cranking of the OODA loop is required, iterating as fast as humanly possible.

2. You must cultivate true sight

The second big lesson, acquired through fear, is that there is a certain boldness that comes only from true sight. Seeing clearly means seeing your weaknesses without stubbornness or pride. Seeing the world in harsh relief. Killing your darling beliefs and philosophies without preciousness. Putting yourself out there repeatedly without fear. Leaning into discomforts – in order to see them clearly, and outgrow them as fast as possible.

3. You must transform yourself

I firmly believe most people are capable of much more than they know – they simply haven’t pushed themselves anywhere near the jagged frontier. Reality is malleable, and your self (which defines your reality) is the most malleable of all.

That’s all for now. Until next time,

Ben Guo | hraness.com

PS) Here are some lyrics to Grey Gardens by Casey MQ that feel especially poignant today:

Subscribe now

In the grey gardens
The deeper I go
Into the crowded room of people I know
In the grey gardens
Where everything grows
It's such a long way home
Falling like Alice did into something so cavernous
Banality becomes something like stars into your heart
Reach into me
Take everything
Remembering is not opposite of forgetting
It's like the lining
Peeling back my skin and leave no traces of my body
When I sleep to drеam my orchestra plays theme of soul and spirit pouring out
Rеach into me
Take everything
Remembering is not opposite of forgetting
It's like the lining

Catching you up: I worked at Stripe for most of my adult life. Then I left to start a company with a friend, (who built Substack). I’m taking a break now, but Zo is still going strong. I decided to move to Puerto Rico with , and venture into the unknown.

I chose the moniker Hraness recently, inspired by the agent harness. It’s really quite astonishing that a simple perceive-act loop can elicit such useful lifelike behaviors from LLMs.

Hraness also means “Ra-ness”. Ra called all forms of life into existence by uttering their secret names. Every evening, Ra journeys into the underworld, transforms into his ram-headed form, and is joined by his buddies Sia (perception), Hu (command), and Heka (magic power). He defeats the snake god of chaos Apophis, and returns at dawn to triumphantly light the day.

I developed a severe case of AI psychosis recently; it was so severe that I had to take a serious break and move to the tropics. Since Fable & Sol, I’ve become convinced that the laws of physics in software have changed, and we must all change our ways. Building a software factory is like tending to a fertile plot of land: gently coaxing agents until the land blooms into a verdant jungle.

I sincerely believe that now is the time to venture out into the latent space as a fearless explorer, savoring that nebulous new air. We’re entering a brave new world in which magician-explorers can shine a light on dark terrain, re-wilding our digital environment into something much more strange and vibrant. Stay tuned, and I’ll do my best to teach you the Way of Hra.

Until next time,

Ben | hraness.com

PS – I’m making music again. Keeping it simple for now, just saw waves and 707. Some loops from this past week:

1. Jungle ⌁ 2. NMQC (Bad Bunny) ⌁ 3. Beach ⌁ 4. Lagoon

Scenes from my life

What I’m consuming

Sunshine, salt water, smoothies, chili, and mamey sapote (h/t Bill & Alexa)

Permutation City by Greg Egan (h/t Ian)

27a Pitfield St by Bassvictim (h/t Ryaan)

The Coward Forgets His Nightmare by Chanel Beads (h/t Dan)