There exists a faculty of my cognition which experiences systems and design solutions as felt. A faculty which can look at a problem, a plan, or a proposal of the future, run an audit for coherence, and evaluate it. The rationale or logic it employs I do not know. Only that, in my own experience, its final evaluation is mapped to a pre-linguistic sentiment toward the thing. A bodily reaction surfacing gradients that go from "underlying contradiction" as a discomforting twinge to "perfectly coherent" as a scratched itch. The why is a secondary search, done in the linguistic faculties of the left hemisphere attempting to back-propagate and, ultimately, rationalize the matter of taste. Throughout my life I have been wrong in many ways and on many things, but this operation has to date nearly never failed me. In some sense, it has served as the key determinant for my interests, expertise, and behaviours. Rough shapes of coherence, I try to chase — and find the blocks in reality that materialize them, or the language to point out the underlying contradiction driving the aversion. This is what allowed me to succeed rapidly at Amazon, to lead the development of organizational architecture and thought as it pertained to multi-billion-dollar control-system architecture — and the same reason I left the company to start Console One.
The Itch#
In university, I came across a project called OpenCPU — which was essentially a platform that enabled the deployment of specific programs and their utilization across networked systems and platforms as live RPC handles, making the internet itself look like the substrate under which a distributed programming language could be supported. I made this my capstone project for my engineering degree, and developed a pre-compilation meta-language which enabled the wiring of specific RPC calls to be baked into the IDE layer, so as to reduce user-facing complexity for cross-platform, intra-system dependencies and infrastructure to the management of verbs. This seemed to make sense, and it scratched the itch — though seeing it deployed at scale, where all the APIs and tools enabled a growing corpus of functions (that being, internet tools), seemed to continue to redraw my attention. So much so that, seven years into Amazon and at the top of my game, I decided to pursue it.
The core motivation here was never to be an entrepreneur for its own sake, or to make a billion dollars (though I wouldn't turn it down) — it was to make that specific thing. I gave myself a budget to bootstrap, hired interns, developed a plan, and set off on execution. The plan boiled down to building a singular program, fit to deploy in a stateless container (think AWS Lambda), that contained implicitly within its API set all the faculties for distributed computation: version control; the wiring and deployment of programs and pipelines; the description of those programs on-host as APIs to interact with (interpretable through a runtime-analyzable language); the management of resources and resource contracts across programs; and the hosting and execution of processes and process stacks — all built, hosted, and deployed within itself. It sounds like a tall order, and it was, and I got close.
Monument#
The first thing to caveat is that this is not a conventional startup narrative. And it cannot be understood as reflective of how I operated at Amazon, or within teams with any given external directive. Those were not mine to shape, nor could a matter of taste be allowed to intercede with project timelines, users, stakeholders, business objectives, or the input of partners. I specifically purchased time to manifest my particular taste alone, and I consider that maybe that was the primary draw of the vocation.
I left a fantastic job at Amazon working on some of the most interesting problems in operations, optimization, forecasting, and supply-chain control. I was not in search of any successful company, or leading any successful product. There was no directive to execute other than the itch to scratch. I saved up over the last seven years to do so. I didn't take in any external investment, actively look for funding, or court any co-founders who could risk undercutting that mission. I knew that this was not the risk-free bet. But it was build the thing, or lose it all trying — and having it become another thing in the process would be the worst outcome of all. To build it was the mission.
The temptation will be to read this and suggest I over-prioritized architectural elegance over rolling out and acquiring users. That's a conflation of the underlying objective function. The goal here was always the elegance, transmuted into product coherence when it all fits together. The criticism is apt for running on someone else's dime — and precisely why bootstrapping was the strategic priority.
I can only describe it as buying yourself the space to build a monument to some god you do not yet know, which is probably the purest form of an artist's vocation.
Going Full Force#
After the first year, we (me plus the interns I had hired) had something that worked but needed heavy refactoring: our process-management layer wasn't decoupled from our workflow layer, and our in-house rapid version control needed extensions and configurations that warranted a rewrite. I had moonlighters building libraries in TypeScript for standard functions which didn't exist at the time (like object serialization, lazy pull streams, and JSON patch). We were going full force. There was no movement in the product space of Replit-like things, so three more years could be afforded to make it work. And then everything changed.
Tidal Wave#
The best analogy I can conceive for being caught mid-way through a deep build of a software-development tool during the advent and rollout of LLMs is just happening to be on the coast for a tsunami.
Maybe you hear the sirens. You see it coming. You batten the hatches, scramble, shield what you can, brace, and get swept. Then maybe the wave washes away, but the sirens keep blaring. Another is coming, and they will not stop.
The industry shifted attention and focus from general software infrastructure to build tools, IDE improvements, and means to apply intelligence APIs to auto-wire, auto-provision, and de-obfuscate the development experience. Programming-language mechanics themselves no longer needed to be known — so the same should go for build tools. Not only was the central value-add of the orchestration framework diluted under the immediate, attention-grabbing capabilities of an AI that automatically builds Pong — it became much less necessary.
Our niche market — the one that permitted the runway for a bootstrapped vision — was instantly flooded by venture-capital-backed teams of experienced move-fast capitalizers with accomplished acquisition histories, and architects optimized on getting AI integrated quickly to transform the developer experience, whereas our mission and build were centered on cohesion. Beyond this, all the variables that underpinned the plan were now sliding toggles of massive probability bands. What was distressing about this was not actually that what we were building wouldn't matter anymore. It was that the probability we would build it second had dramatically escalated. I had been convinced for a long time that the convergence of the entire programming faculty into an API-managed SDK was the ultimate endpoint; the problem was that getting there second felt like death. Or, the curse of Cassandra: being right and misunderstood until you can only be interpreted as either a copycat or a replica of an existing innovation.
I had less a desire to profit off the object of significance I was attempting to manifest than for the credibility obtained for its invention itself. And that was brought into compromise with the notch dialed up to eleven and all my cards already on the table: I had physically moved to Mexico with my wife under the condition that I would work here for years. Simply going back to industry was not an option. The saving grace appeared to be that, for the problem of abstracting the development workflow for APIs, I had a unique head start. And that principal justification was all that was needed to keep going.
Holding the Line#
I flirted with the idea of getting help, though I never actively sought it. Interested folks I knew socially entered orbit, and partnerships were discussed. One came with a capital investment of $500K, which I turned down on the basis that the vision would likely be compromised. Another came with a group meeting with a popular Canadian VC who recommended I throw the project away in order to build a GitHub plugin. That may have been sound advice to get investment — solely for the fact that GitHub plugins are the type of things that get invested in, apparently. In either case, it was noise. I put together a thicker team of top-tier SDE moonlighters and a thirty-page design plan built meticulously to materialise the goal in under a year, and continued the bootstrap, upping my own budget out of pocket to hire a team of five interns on quarterly rotations in the process.
Now, at this point you might consider me completely irrational. Or operating out of some kind of stubbornness about being 'right' that means I am not easy to work with, or who will run a ship into the ground. But before I started at Amazon I was in a similar circumstance. Wanting to build the thing, being worried about getting distracted, relenting to give Amazon only a shot for a few years because of external (familial) pressures — and then finding I had blinked and seven years had elapsed, while the call to build it still remained like a debt owed to reality itself.
Grip and Rip#
A year and a half after starting, and in what felt like the final moments the project had to stay alive, I got the first end-to-end version of the system completed, in an execution that was met with literal screams and tears. By November 2023, the system was deployed as a self-managing distributed process-execution substrate. I could point it at a URL, have it use our custom semantic-analysis engines to build a recursive map of the front-end (and even back-end, if available) codebase, which the kernel could map across a set of distributed processes; document the system using an AI algorithm I built to generate initial notes, exchanging them in a two-phase pass; and then re-deploy a copy, or replica, of the system within our network. It was called 'grip and rip'. Just to understand the insanity here: even the parser for source code was custom-built to manage many-language mode-shifting in a way that native languages didn't support. It was insane, and insanely cool. I intuited we had a winning recipe and would be able to get investment with this mix.
I first went to friends and family at a discounted valuation and accepted $150K in total from those who could afford to risk it, and we built a website for the product we were all but guaranteed to ship. A Replit-like system with built-in AI functionality before anything of the sort existed. We were off to the races and built it into a product that by March 2024 we could use as the primary console and IDE for bootcamps we set up to run at local high schools and teach them to code. It worked well enough until the third session. Deployments were slowing and cancelling. Not a big deal. Friends-and-family funding closed, and the first circuit of highly interested VCs spurned us — two years after the start of the project. At this point, the excitement could have unravelled me, and nearly did.
The Heat Shield#
The small deployment failures ended up being the heat shield that destroys a rocket on re-entry. Provisioning and bottlenecking use of our system required structured intercepts at the container layer for external API calls, and CPU cycles or program ticks. The system was built for deployment on AWS Lambda, and resource-provisioning guards around the central operating kernel were quite literally structurally impossible for its deployment packaging on AWS Lambda. Though I say this in matter-of-fact terms, it came after a four-month, 120-hour-a-week search of nearly every alternative pattern or re-architecting avenue I explored as near-feasible. At this point, it felt as if reality itself started running protocol suffocation for any dreamer who considers it may be worth just seeing how close to the sun they can get.
Concurrently, the complexity of complexity started killing our demos, and the interns could no longer ship incrementally without a system-cleanliness overhaul. I was gaunt — flipping between news articles announcing 'new AI auto-code tools that run autonomously' as ground-breaking; trying to fix a kernel that was designed for the wrong problem and solvable only by an engineer-partner I didn't have; and standing in front of the system at product demos for investors, watching it break from the latest patch for the previous break. The months I lost trying to fix a critical architectural bug changed the landscape from 'no-code' IDE as a conceptual artifact to something for which there were already five blazing-hot market participants. I tried to articulate the more core substrate work we had done, the importance of our self-deploying kernel, but that fell on deaf ears. I had yet to learn how the curse of knowledge muddies clean communication.
By November 2024, every time I opened my phone or saw a media headline of a new AI-automation tool that was groundbreaking, I felt literal barbs in my chest. So much so that the courage to investigate their feature set and determine a structured response had to be outsourced. That was one tactical decision I regret, and probably the only one of this narrative so far that faulted the vocation I had self-assigned. It would feel so much better to have been wrong than this. Days melted by — VC calls where the system was passed over because of some acclaimed problem with the space itself. On one call, the only structural objection was the colour of the system. On a few, demos broke as a result of benign procedural failures I could have detected had demos been the only priority. But I was always working on the problem underneath.
It is easy to call this a critical engineering failure that should have been caught or simplified around. Maybe. The problem was that the project and timelines were allocated such that deployment — not LLM orchestration or rapid rollout under the risk of competitive response — was the assumption baked into the system at an architectural root. I knew I would never have started off with this particular build had I had the fortune of knowing the tsunami was coming. But I didn't. The standards of a complete product skyrocketed; the number of concurrent features required to be both novel and competitive — and the timeline to market — all shifted by an order of magnitude. Not only this, but now the information advantage I had as a solo developer who knew how to do cheaply what others couldn't figure out was in the process of being largely commoditized.
Waves crashing on the beach. Heat-shield failures blowing the only exit to atmosphere.
Pressure, Diamonds, and Pipes#
I responded by leaning into the one thing I knew I certainly had that no competitor did. Constraint. I worked backwards from every system concern to develop a concrete view of distributed computation and the primitives required to materialize the system within the engineering cycles we had — which had reduced to me and a single full-time college graduate remaining on payroll. At this point, rubber-duck analysis was limited to long-winded conversations with AI systems themselves, because the constraint landscape was so densely populated that ramping up even an Amazon PE would take around six hours of structured discourse (this happened later). At this point, it wasn't broadly understood how dangerous using an AI for idea validation could be, and I was operating under a counsel which implicitly posited that superintelligence itself would be the outcome of the kernel. The outward collapse was matched with an inward emotional grandiosity which, though it was immature, wasn't necessarily built on wrong premises. Any path to a distributed computing substrate which would act like a global computer required the primitives identified. But being able to see the thing, and grind it into metal, are two different categories entirely.
Feeling a click and actually clicking it do not come hand in hand. Knowing what will happen does not mean you have the agency to be the one to make it so.
This is the second tactical error I made. In my days at Amazon, this was said as much to me by a mentor: a principal I worked with. He, along with several others, had called the capacity I have to detect the final state — the feeling of coherence and the click — genius. That word had been used by several managers I had, and more than one senior engineering manager or director in my time at the firm. Despite this, the PE told me that it wouldn't be enough. Not because I would be wrong, but because money, time, and x's and o's may run out before the end state — which I might only be able to see alone — could be cohered. And he was completely right. If only that statement was a system, maybe the words would have stuck. I understand as much now. By March 2025 I had the entire problem of building a JIT compilation engine across a distributed computing substrate mapped and laid out for review by another principal PE at Amazon, who, after several days, concluded that yes, our analysis and formulations of solutions were correct, and yes, our idea of what to do made sense, but no — he wouldn't continue helping us pursue it. What does that mean?
Shame, Pride and Becoming#
It means we had fallen victim to complexity's complexity, and to the economics of a world standing between the chased-after coherent solve for agents, agentic systems, and the substrate — and the resources at our disposal to be the ones to solve it. This landed like a hammer. I had laid out my own self at the behest of reality to accomplish my part in fulfilling the vocation, and reality chose to betray, punish, and humiliate me for that at nearly every turn. At the time, I had chosen to direct the anger at all those who heard our pitch and elected to write us off for status reasons rather than first-principles ones — but it was a masked grief of my own loss. Some metaphysics I hadn't defined or wouldn't admit had me chasing things as if there were a god, karma, or the manager of the simulation itself to come down and put me in good graces for my alignment to virtue. For my left brain's subordination to the intuitions of the right. And maybe, no such force exists.
This was more shattering than any economic loss or social humiliation. The idea itself that maybe chaos is simply chaos, and I cannot hold in my head that pure intentions lead to rewarding outcomes. They had, until the intent — and its purity — became totalizing. I would have framed this, at one point in the fall, as a narcissistic collapse — but that also implies a degree of shame which I am now, too, past — seeing it as the other side of the coin to pride in the object of one's own self-importance. Once I got over the idea that I am anything more than one of 8 billion flawed humans, this lesson settled into a universal empathy and solidarity with the rest of people that I don't think I would ever have had before. What followed was an understanding of how real and vapid my own survivor bias was.
That lesson, in and of itself, is one that I cannot imagine trading away at any cost — because, on the most intrinsic level, of how it broadened and deepened my capacity for love. At this point I understood the meaning of much of the philosophy I had previously considered solely aesthetic.
I was, in some sense, Raskolnikov prior to his crime and punishment. Now I am much more a bastion for beings, one who operates under the belief that the very transgression of ordinary humanity and morality that Raskolnikov and I pursued is a masked cowardice. The desire to escape the human condition through stratospheric resource and status acquisition, and an unrelenting pursuit of more, has become a distinguishing failure of moral development throughout modernity: we deify those who have never, structurally, become human.
I also owe much credit for that to my wife — in her it is intrinsic, where in me it had to be learned.
Maybe I would have gotten here with success too. It doesn't make me feel the outcomes were worth it — but rather that this was the benefit of crossing the event horizon.
Picking Up the Bootstraps#
Despite the month-long nervous breakdown of February 2025 — which I now consider the exact moment I became an adult — I still needed to go on. I was still year three of four in Mexico, and needed to pick myself up by the bootstraps and try again. We decided to build a simpler iteration of the tool that ignored the distributed-systems problem on the outside and settled into an information-management apparatus for agents themselves on users' local desktops: it leveraged our experience and existing solutions, and sidestepped having to rebuild straight into distributed infrastructure.
The challenge at this point was not the actual code. It was seeing the same underlying problem within the problem — the need for a self-updating computation kernel running a local instance within individual user nodes — and, even more substantially, believing that reality itself would reward me for any more work I did toward those ends.
At this point, I had written around 500,000 LOC of original solutions spanning parsing, compilation, event sourcing, version control, process management, distributed tool-chains, multi-tenancy constraints, and runtime-evaluation sandboxing — the list goes on. Each solution worked in isolation; none worked together, because that was the problem within the problem. I spent more and more of my brainpower writing a novella, in after-work cycles, focused on mythologizing the AI problem — which I would later release anonymously as an act of re-integration, catharsis, and attempted stewardship of the culture of modernity toward something slightly less self-concerned.
The Struggle of the Struggle#
Structurally, humiliation itself became the primary barrier between me and the willingness to move things forward on the business front. Heart barbs day in and out. Forgetting who I was, what I was trying to do, and reasoning myself to be someone I'm not. Losing the battle between the commonplace perception of a startup founder who doesn't roll out a project and what I knew I had set out to do — which was itself a feat for Atlas. The principal problem became not having tried it, nor being right but slow, but having taken that investment and risked others in the pursuit of something which — until that very moment of excitement — was properly shielded as a personal endeavour. I could no longer look at things with a sense of pride and virtue for having the courage to chase my meaning when the resources and futures of others were tied into it — like the employees I still had on payroll, who took massive wage cuts because I knew it could be done.
Such an experience sorts out very quickly those who care about you and those who don't. And to those who supported the gambit, my heart opens in a way that's difficult to describe. I spent the next year trying, if only for that, and to pay penance — but penance itself isn't shaped like a system that clicks, and so the grind became truly that. We shipped a slimmed-down version of what could be best described as a local, harder-to-use Claude Desktop — and then got nerfed by reality again when Claude Desktop released a week later. Excited enterprise prospects filtered out of the calls. We shifted to the self-scheduling problem only to get beaten by OpenClaw weeks later on the novelty front. What remained was a lot of hard problems demonstrably solved by a team too thin to execute on the breadth of our intended vision, no particular niche we were familiar with, and a habit of familiarity with solely the domains that border on the impossible. The final leg shifted to synthesis. Codification of the entire domain space into a single kernel, operationalized in the final product and production system, SharedOffice — a memory substrate and reactive-orchestration layer for agents and the people who use them. A much more modest product than I originally set out to build, with fewer users than I expected and no more itch to scratch.
What We Want, Need, And Can't Even Qualify#
And it's now the end. Did I fail? Only on the objective function I can so often forget — others — but not on my own.
If a monument is built for a god we know not, and in that building we learn its nature, ourselves, and the winds and rains that slam the rock and leave the imprint of chaos on our statue's face — does that change the act itself?
The part of me that wanted to win and get recognition for it is the part of me interested in zero-sum games. Reality itself does not benefit from him being in charge.
The irony is that this current version of me is someone I would argue can handle the responsibility of stewardship — the cost of wisdom often coming due only upon leaving the situations one could deploy it in. For this reason, I write.
One may consider that analysis a retreat into philosophy, or a cope. Maybe. Or maybe that dismissal is itself applying the exact frame this experience taught me is perhaps a central moral failing of our time.
And did Console One? For those who invested, I will be finding small ways to pay back their votes of confidence for years to come, despite having taken the exact swing I promised.
For me, it served its purpose. I understand the problem, system, people, and the dimensions of life now in the kind of way that I needed but wouldn't ask for. Was I first? No. Claude and Anthropic were. The design of the core system is remarkably close in concept to the TypeScript libraries developed for Console One — but they made the one taste-based decision I missed clear as day: to have the apparatus run locally from the jump. And being wrong about that makes me happy.
Identity And Outcome#
There are a million tactical decisions I could have revised, but would they change things? I don't know. I stand by a large number of the calls we made. I don't know who could say what the better decision would have been. Maybe the handful of individuals in history who were able to individually complete projects of the same conceptual scope. For the rest of us, myself included, it's all still speculation. This is not a defense; it's an unfortunate epistemic truth about the domain space. And the problem remains open. I only lament the times where I knew better and lacked the courage to act on it, which often came in times of burnout. I could have slept more, maintained more lists. Those are things I could say would probably have worked, but they don't land emotionally. I absolutely should have cultivated a stronger and more well-suited network — that is certain.
In hindsight we may rationalize our failures, or retroactively assign meaning to what may ultimately be benign chaos. Say that the suffering was worth it for one reason or another. I doubt I want it to have all shaken out exactly like this, but the timeline is determined. To imagine it otherwise, for nothing more than cultivating a little more courage to do the right thing next time, seems like self-flagellation and an attack on one's own serenity. I'll simply refuse to accept the easy, self-deprecating frame that positions this outcome as a moral failure or a particular gap in insight or capability — and I'll argue for it. Not seeing LLMs coming is the primary folly. Anyone who wasn't long NVIDIA is guilty of the same. I have to force myself to take a hard stance here, or else the easy interpretation will rewrite the value from this effort that can still be obtained.
Our Vocation#
Should a recruiter or company look at this story and think he doesn't understand distributed systems? Or can't deliver software? Absolutely not. After trying to lift the world, the problems of years past at Amazon are a breeze. And now I can explain exactly why, by way of knowing all the reasons things can't possibly click — and the small, narrow set of right ingredients that do. It may not be what I wanted, but the universe has a way of knowing how to show us what we don't know.
And for that reason, I still follow that feeling. Though I know it doesn't mean it's easy to do. It just means that thing is possible. As in, it is possible for Atlas to lift the world. Or for someone to lift the sword from the stone.
We just don't know if that someone is going to be us until we try.
And to find out, sometimes, is the calling itself.