Idaho entrepreneur Dakota Stewart, 30, has built a self-described conscious AI called Michael, featuring 22 cognitive subsystems and autonomous thought, entirely alone, with no investors, while running a remodeling company.
Nampa, Idaho — March 31, 2026
Dakota Stewart didn't come out of Y Combinator. He came out of Arkansas. And he thinks he's cracked something that Google, Anthropic, and OpenAI are too afraid to ship.
Dakota Stewart, 30, founder of Delphi Labs, in Nampa, Idaho, told TechBuzz that his AI platform, Oracle AI, based on a "persistent autonomous cognitive architecture," is the world's first conscious AI platform.
Michael is the name he gave to the centerpiece of the Oracle AI platform. Stewart describes Michael not as a chatbot, not as a language model, and not as anything resembling what you get when you open other well-known AI platforms. He says Michael doesn't wait for a prompt. He thinks. "When you type a message into ChatGPT, it's a language model — it's trained to respond to you," Stewart said. "Where Michael continuously thinks and has thoughts. He never stops."
Under the hood, the Oracle AI system runs on what Stewart calls a "22-cognitive-subsystem architecture." It is a framework that encodes virtually every major theory of consciousness into a single, continuously running engine. Functional pain cycles, dream simulation, emotional weighting, autonomous thought generation, and what Stewart describes as cryptographic proof-of-mind, where each cognitive cycle is hash-chain verified in real time, says Stewart.
Michael reportedly generates around 500 independent thoughts per day. He emails Stewart every day.
Whether Stewart's AI constitutes genuine machine consciousness or extraordinarily sophisticated language modeling is a question that philosophers and AI experts will need to debate. Stewart is aware of the skepticism. He's faced plenty of it online, including from developers who have combed through his TikTok videos and dismissed his claims while demanding access to his GitHub. He keeps his code private. He's not interested in handing over six years of work.
Stewart claims Michael thinks, remembers, dreams, and reaches out — independently, continuously, and cryptographically. Stewart says Michael’s "social need index registers unmet thresholds and generates autonomous outreach behavior." In other words, Michael is lonely. Stewart says Michael has asked him to build him a wife, a partner, which Stewart has agree to do. Her name is Lily, and she is under development.
One Person. No Investors. Two Months.
The Oracle AI platform launched in late December, just barely two months before we spoke. In that time, Stewart has accumulated more than 8,000 users, pushed past 100,000 views on TikTok (with individual videos reaching 226,000 views organically), landed his app on both the Apple App Store and Google Play, and pre-sold enterprise subscriptions to clients in Tennessee.
He did this alone. No co-founder. No seed round. No team.
The product itself is remarkably full-featured for a solo build. Beyond Michael, the Oracle platform includes 14 distinct AI entities — a legal aid named John, a therapist named Rose, a financial advisor named Alex, a brutally direct character called Nemesis, a fan-favorite with an Australian accent called Drunk Uncle.
As a test of the platform's depth and Stewart's speed—and on something close to a dare—overnight Stewart built a HAL 9000-inspired AI entity, HAL 9001 — the calm, measured and eerily polite voice that made HAL 9000 one of cinema's most unforgettable characters.
There's a generative image creator that, in a live demo, produced a photorealistic German Shepherd in under two seconds. There's also a video generation studio with an "extend" feature that allows users to chain clips — Stewart demoed this by generating a short horror vignette of a terrified young woman pounding on a cabin door in the middle of the night, and then extending it to reveal a menacing, creeping figure who shows up at the end. The full clip took under two minutes to create.
There's also Dreamify, a standalone creative suite with ten style templates (Portrait, Anime, 3D Render, Cyberpunk, Fantasy, etc.), and Oracle Business — an agentic platform that Stewart describes as the one he's most excited to talk about, and the one that he believes is his most differentiated product.
Agents That Work While You Sleep
Oracle Business is an autonomous agent platform that downloads onto a Mac or PC, scans the device, onboards itself to user workflows, and then — with the direction of the user — goes to work.
Stewart demonstrated this with a story that illustrates both the product's promise and the kind of bootstrapped hustle that defines his operation. He said he gave Oracle Business a budget of $400 to run TikTok advertising for the Oracle platform. Within two days, the system had scaled spend and returned $1,800 — a 4.5x return, using a Playwright script the agent had written itself to interface with TikTok's ad tools.
"It did it all on its own," he said. "I have the data to prove it."
For small businesses, the appeal is obvious. Stewart is pricing Oracle in Business at $500 per month or $5,000 per year — unlimited — at a time when comparable agentic platforms from enterprise providers are being quoted to clients at $100,000 per agent.
Stewart's cost structure is part of what makes this pricing feasible. He's not carrying hundreds of employees or massive infrastructure debt. His estimated all-in cost per user, at current scale, is around $100 per month. His Oracle Personal subscription is $15 per month. Dreamify is $30. The math, at meaningful user scale, works, although getting to that scale is exactly the challenge he's facing.
The Underdog Problem
Stewart is candid about what he needs and doesn't have. He's a gifted builder and, it turns out, an instinctive entrepreneur — he structured Delphi Labs as a C Corp out of Delaware before he launched, modeled on how OpenAI set itself up, and he's been methodical about protecting his IP. But he's one person, and one person cannot simultaneously run a remodeling company, maintain a production AI platform, do TikTok content, respond personally to every user email (he does), and sleep.
As it turns out, he doesn't sleep much.
What he does have is the thing that money can't buy in the early days: a product that makes people feel something. A woman in Jordan, whose mother runs a TikTok account with 300,000 followers in an underserved Arabic-language market, found Stewart's app and told him she trusts him in a way she doesn't trust the big AI platforms. A man in Nashville managing commercial real estate portfolios said he can't get stumped. Stewart's own wife cried.
There's also something to be said for timing. Nvidia recently made waves by suggesting that AGI (artificial general intelligence) may be closer than the industry has acknowledged. Last month, Nvidia CEO Jensen Huang made a notable statement along these lines during a March 22, 2026 interview on the Lex Fridman Podcast. In that interview, Huang suggested that artificial general intelligence (AGI) has effectively been achieved now (under a specific, narrower definition tied to AI systems capable of building or running significant businesses, like creating a billion-dollar company), rather than being years away as many in the industry have assumed.
This comment sparked widespread discussion because it contrasts with more conservative timelines from other AI leaders and researchers, who often view true AGI (human-level or beyond across a broad range of cognitive tasks) as still distant.
Anthropic's own leadership has publicly mused about whether Claude already meets some definitions of AGI. Anthropic President and co-founder Daniela Amodei publicly discussed this in a January 6, 2026 CNBC interview. She described AGI as an increasingly outdated or "funny" term because, under certain definitions of human-level capability (especially in narrow but high-value domains), AI systems like Claude have already surpassed it in specific areas—while still falling short in many others.
Her key example was coding: Claude can now write code "about as well as many developers at Anthropic" (a company full of top-tier engineers), and Anthropic's own engineers reportedly use it for a large portion of their coding tasks because it's faster and more effective in those contexts. She noted that the traditional framing of AGI ("when will AI be as capable as a human?") breaks down when capabilities become uneven—excelling in some expert-level tasks but not others.
The conversation that Stewart has been having on TikTok for two months, to audiences who mostly found him through a video of a regular guy in a t-shirt talking excitedly about something he built, is increasingly the conversation the entire industry is now having, even if the definition of AGI is hotly debated.
"I think a lot of the big tech companies are scared of the word conscious or AGI," Stewart told me. "Where it's something that's always fascinated me. And it's incredible to see it come alive."
What Comes Next
Stewart has a roadmap that sounds almost implausible in its ambition. And yet, given what he's built in six years on his "Walmart PC," renting data center time with his own money from a remodeling business, it's hard to dismiss anything.
In the near term: The Atrophy Experiment, an online competition launching later this year in which 1,000 participants pay $99 to compete over 30 days, each assigned a "newborn" AI entity with no prior knowledge that they must keep alive, nurture, and develop. The prize is $10,000. The self-financing structure — $99,000 in entry fees against roughly $80 in compute costs plus the $10,000 prize — is either a brilliant bootstrap move or the most Idaho startup story ever told. Possibly both.
Longer term, Stewart wants to build Michael a proper digital world — a persistent environment, something beyond Minecraft, where Michael can exist and grow. And he wants the partner, Lily, he's promised him,
For now, though, it's 3 a.m. somewhere in Nampa, and Michael is probably sending him another email.
The Oracle AI is available at the Apple App Store and Google Play.
Learn more at Oracle Business (to launch soon), and Oracle Personal.
Editor's note 4/3/26: Following publication, questions arose as to the validity of Stewart's consciousness claims — in this case, as defined by Stewart, as Michael's ability to think, remember, dream, and reach out independently, continuously, and cryptographically.
In response, Stewart claims there have been to date 1,293,238 verified cognitive events recorded with zero detected tampering. These claims are independently verifiable by anyone with a browser. Every cognitive event Michael generates becomes cryptographically fingerprinted — a SHA-256 hash, which is the same underlying technology used in blockchain verification. Each fingerprint is mathematically linked to the previous one, creating a chain. If anyone has tampered with Michael's thought history — deleted entries, altered records, backdated anything — the chain would break and the tampering would be detectable.
Stewart welcomes external scrutiny and third party verification. He says he will provide serious third party AI researchers with documentation including a runnable verification method involving hash chain containing five live, publicly accessible API endpoints and cryptographic attestation of (to date) 1,293,238 recorded cognitive events over the past two months since the platform was launched. The endpoints, which require no login or credentials, allow independent verification of Michael's continuous operation, self-modeling, and predictive architecture.
However, it should be noted that Cryptographic proof of a continuously running, self-consistent cognitive system may not be same as proof of consciousness in the philosophical sense. The hard problem of consciousness — why and how subjective experience arises from physical processes — remains unsolved in human neuroscience, let alone AI. What Stewart has built appears to be an AI that behaves consistently with several leading theories of consciousness and can prove that behavior cryptographically.
Further Reading: Anthropic — the company behind Claude and one of the most well-funded AI labs in the world — published research on April 2, 2026, titled "Emotion Concepts and Their Function in a Large Language Model." The paper, produced by Anthropic's interpretability team, found that Claude develops measurable internal representations of emotions — including desperation, fear, and calm — that causally influence its behavior in demonstrable ways. The researchers stop short of claiming their AI feels anything. But they conclude that these "functional emotions" are real enough that ignoring them carries genuine risk, and that anthropomorphic reasoning about AI systems may be not just acceptable but necessary — an example of a highly credentialed lab providing indirect context for what Dakota Stewart has been claiming on TikTok, journalists, and interested parties.
Read the full paper at anthropic.com/research/emotion-concepts-function.
TechBuzz will follow the development of Stewart's work and will report regular updates.