The clue was not buried in a pitch deck. It was lying among the metal chips. Justin So spent his summers at EDS Manufacturing, his father's machine shop, doing the useful work entrusted to a young pair of hands: deburring parts, sweeping the floor, watching jobs move through Haas and Doosan machines. He also watched information fail to move. A procedure sat in a binder. A work order lived in one system, machine data in another. The best answer was often lodged in an operator's memory, available only if that person happened to be nearby.
Factories manufacture precision, yet their knowledge can travel by scavenger hunt. When a machine stops, the answer may already exist, but existence is a poor substitute for arrival. Someone has to find the right page, the right login or the right veteran. Meanwhile, a line measured in seconds waits on an answer measured in footsteps.
So carried that mismatch through an economics degree at UC Santa Barbara, a master's in computer science at the University of Pennsylvania and a finance role at chip designer Arm. In 2025, he co-founded Sidekick. By the summer of 2026, the San Francisco company was in Y Combinator's S26 batch. Its proposition sounds almost impolite in a software industry fond of adding destinations: do not give the worker another place to go.
“Nobody on a factory floor wants another app.”Justin So, on Sidekick's interface
The interface that was already installed
Sidekick lives where the conversation already does. A worker texts a question by SMS or a familiar messaging service. They can send a photograph when a machine is difficult to name, or a voice memo when gloves and grime make typing absurd. Sidekick looks through the company's own procedures and records, answers with that context, and can open and route a work order. If it cannot answer, it escalates the question to a manager. The reply is saved so the same problem has a shorter journey next time.
The worker-facing simplicity hides the more ambitious product. Sidekick connects machine readings, fault feeds, documents and operating history. It is meant to notice trouble as well as receive it: a drifting measurement can trigger a message to the right person before the machine becomes an expensive statue. The front end remains a thread, a small doorway into a rather large room.
There is discipline in choosing such an ordinary interface. Enterprise products often treat adoption as a training problem: if workers will only sit through the tutorial, create the account and remember the password, the software will reveal its worth. So reverses the burden. The product should learn the workplace's language and habits. A worker should not have to become a minor software administrator before asking why Machine 3 dislikes cold mornings.
A family shop becomes the first customer
EDS Manufacturing was more than the birthplace of the observation. It became Sidekick's first customer. That is a neat founder-story circle, but also a demanding product test. Family history does not make a factory less noisy, a machine less temperamental or an operator more eager to humor a half-ready tool. The software had to be useful where So had first seen the problem, among equipment he could name precisely: a Haas VF-8, a Doosan DNM 5700 and a VF-4SS loaded with five vises.
His public descriptions linger on concrete details. One alarm code should stop everything. One machine needs a warmup cycle on cold mornings. These are not the kind of truths that make a handsome corporate handbook. They are conditional, local and earned through repetition. They are also exactly what disappears when an experienced worker retires or simply takes the day off.
Sidekick's answer is to capture knowledge during the act of using it. When a manager replies to an escalation, the answer is not merely consumed; it becomes part of what the plant can retrieve later. Work writes the manual. The manual, unlike a pristine binder, has fingerprints.
The numbers, and what they do not explain
In August, So said Sidekick had passed $38,000 in monthly recurring revenue, twice the previous month's figure. He also reported it live in four countries, serving machine shops, plastics manufacturers, metal finishers and contract manufacturers. Another company update put the footprint above 20 plants and 600 frontline workers. These are self-reported early-company numbers, snapshots rather than marble inscriptions. They nonetheless suggest that the plain-text entrance is finding doors.
The more interesting measure may be time-to-trust. Factory technology competes with scar tissue. Workers have seen tablets gather dust and software rollouts arrive with enthusiasm from people who do not work the shift. Sidekick's low-friction pitch is really a social one: do what you already do, and let the system prove itself inside the conversation.
“No app to learn. No tablet gathering dust. Just a text.”Justin So, describing deployment at EDS Manufacturing
There are, of course, serious stakes hiding behind that ease. Advice on a factory floor must be grounded, current and traceable. Sidekick says its answers cite a document, page and revision or the supervisor who confirmed the fix. Its public material emphasizes escalation when the system does not know. The charming part is the text bubble; the necessary part is knowing where the bubble got its confidence.
An economist's loop, an engineer's plumbing
So's route into the company makes more sense in combination than it does as a list. Economics trains attention on incentives and constraints. Finance at Arm offers a view of how technical businesses turn resources into decisions. Computer science supplies the ability to build the machinery. The factory supplied the problem, stubborn and untheoretical.
The result is a product built around a compounding loop. Each plant begins with documents and the expertise already present. Each question exposes a gap. Each resolved gap can improve the next answer inside that company. Across multiple sites, a fix learned in one location can fill a hole in another, while the knowledge remains the customer's own. The software does not merely search the business; it helps the business write itself down.
That aspiration is larger than fixing a bearing. Sidekick's site describes creating training modules and generating documents from captured knowledge. The same thread that answers today's question can help onboard tomorrow's worker. The humble message becomes an input to institutional memory, which is an imposing phrase for remembering what Devin on nights did last Tuesday.
The factory speaks in more than forms
So also frames the access problem as one of language and format. A report typed neatly into a portal suits a desk. It is less natural beside a running mill, on a loading dock or at a station where a worker's hands are occupied. Sidekick accepts the rougher grammar of the job itself: a quick sentence, an equipment photo, a voice recording. Its public examples include workers using SMS, WhatsApp, Telegram and KakaoTalk, with questions arriving in different languages. The channel changes; the company's knowledge remains the reference point.
That distinction matters. Translation alone does not make an answer useful if it comes from a generic model with no memory of the machine in front of the worker. Sidekick's pitch joins two kinds of context: the language a person chooses and the information a particular plant owns. A photograph of a CNC mill is not treated as an invitation for an internet lecture on milling. It is a route toward the manual, prior work and live signals associated with that equipment.
So's posts repeatedly return to the same few phrases - no app, no login, no training. Repetition in startup writing can become wallpaper, but here it also functions as a product test. Every new capability has to survive the constraint. Proactive monitoring, document search, work-order routing and onboarding may be complex behind the scenes; none is allowed to turn the worker's side into a cockpit. The company can add machinery without adding another dashboard to the floor.
This gives the four-person company a sharply defined opponent. Sidekick is not competing only with another software vendor. It is competing with the delay between a question and the person who knows, with the reluctance to document a workaround after a long shift, and with the perfectly reasonable refusal to adopt one more corporate tool. Those are human frictions, old enough to have survived several generations of enterprise software. So's approach is to take them seriously as product requirements.
The lesson hiding in the login screen
There is something broadly useful in So's choice. Builders often begin with what technology can now perform. He began with what a worker should no longer have to perform: a search through folders, a walk to the office, a login to an unfamiliar system, a repeated question to an already busy supervisor. Subtraction set the product brief.
That does not make Sidekick small. It makes its ambition less visible. The company is trying to connect live equipment, written procedure and human judgment without making the person on the floor contemplate the architecture. A well-made wrench does not ask to be admired before it turns the bolt. Sidekick would like to be judged by similarly practical standards.
So's story remains early. Sidekick was founded in 2025; its YC batch is still fresh; its reported revenue and deployment figures can change as quickly as a shift board. Yet the founding image is already crisp: a son sweeping metal chips while information refuses to travel. Years later, his proposed remedy arrives with a small vibration in a worker's pocket. The manual has finally learned to text back.