A platform for creating, running and managing AI employees.

AMTECH AI, Dalton, Pennsylvania. Pre-seed, 2026. Benjamin Palaskas, founder.

A platform for creating, running and managing AI employees.

What it is

Each AI employee is a persistent team of AI agents that does a business’s real work across its tools, files, systems and customers, instead of answering questions. It lives on a computer of its own, with the business’s records, memory and schedules. The owner talks to one agent, which starts each job in its own working session and runs several at once. Between them they browse, research, write code, make documents, pictures and video, send invoices, and keep working after the owner closes the app.

Agency inside, authority outside

The agents are free to work however they need to inside their own computer. Anything that leaves it, such as a message to a customer, a published page or a payment, waits for a person the business has authorized, and code the agents cannot argue with decides who may approve and who may spend.

Who it is for

First, the businesses I have been building for: service businesses and others turning over $1 million to $10 million a year, where a well-configured team of agents can run essentially everything and only the physical work needs a person. And the new ones: since early 2025, applications for one-person companies are up more than 20% while companies that plan to hire are flat (Nasdaq, from Census data), people whose idea needed an employee or two to run.

The deals

In October four businesses signed.

• A Ventura County service company signed for 8% of all its revenue. Its employee does everything in the business except the physical work.
• A Houston moving company signed at $1,200 to set up, then usage. Its employee is being set up to run the whole business.
• A Pennsylvania festival organizer signed at $1,200 plus 10% of all the money the festival raises.
• A crypto trading desk signed at $1,200 to set up, then usage, for an employee built for trading.

How it makes money

What AMTECH sells is not a software seat. It is a growing share of the work businesses now pay employees, contractors, agencies and outsourced services to do. The terms differ on purpose: setup and usage at about an 85% margin, a share of revenue, results-based pricing and monthly plans run side by side, and every request the agents make goes through AMTECH’s own meter, so the data will show which one owners keep.

What is under it

I built the platform every employee runs on: each business’s own records, kept current by its employee; authority in code over who may approve work and spend money; a gateway that meters every request and sends it to the best model from any lab; and a factory that builds employees from one standard and hands them out live, so a visitor to amtechai.com/contractors gets a working employee at the first word. The agents run on Hermes, the open-source agent runtime. What a competitor cannot buy is what each employee learns about its own business, and that grows with every job.

Why now

This fall the largest companies in AI bet on agents that do real work. Avoca, AI for home-service trades, was valued at $1 billion in April, and Hone raised a seed at a reported $285 million in October. Meta launched Muse for Small Business and xAI launched Grok Team Bots. The labs serve companies by sending engineers in: Anthropic’s new services firm says mid-sized companies “lack the in-house resources to build and run frontier deployments”, and OpenAI built a Deployment Company on the same model. An engineer per company will never pay for itself at a small business, and America has 36.2 million of them, with almost half of its private-sector jobs. AMTECH turns that work into a product.

Where it goes

Next, several employees working together for one business; then employees trusted with money inside limits the owner sets; and in time, a way for anyone to start and run a business with a team of agents behind them. Every step is the same platform, with more agents doing more of the work.

The founder

I built the platform alone in 54 days, and I run AMTECH with AI agents; AMTECH runs its own business on one of its employees. I already work full time on AMTECH, about 40 hours a week beside a part-time job, and when we raise I will commit every second of my life to it.

The raise

$600,000 on a post-money SAFE at a $6 million cap. It pays for the computers and the AI the employees run on, free trials, getting owners to the page, and a reserve. I run the meetings and the go-to-market myself with AI agents, so the money goes to the work. At the end of eighteen months any owner who finds a page can sign up without me.

ben@amtechai.com · 805-886-9173 · amtechai.com/investors

AMTECH AI, Dalton, Pennsylvania. How the AI employee is built. Benjamin Palaskas, founder.

AMTECH is a platform for creating, running and managing AI employees. Each employee is a persistent team of AI agents with a computer of its own, doing a business’s real work. This page says what is under it, what I built myself, and what the experiments so far have measured.

1. The employee is a team of agents on its own computer. The owner talks to one agent, the lead, which knows everything running and starts each job in its own working session, several at once. Each session has a browser, a terminal, code, files, computer use and vision, memory that survives between jobs, scheduled routines, and helper agents it can split work across. They keep working after the owner closes the app. Underneath is Hermes, Nous Research’s open-source agent runtime, under the MIT licence; everything above it is AMTECH’s. I fork it at v0.22.2 and own it from there; the employee image changes no line of its core today, so the fork starts identical.

2. The brain standard. Every employee carries the business’s own brain: its prices, its customers, its rules, its past jobs and every decision, as plain files the owner can read and correct. The employee keeps the brain current itself, so one business’s employee gets better every week instead of starting over. The same standard runs AMTECH’s own business, and has since August, with about 600 recorded lessons and every job on the record. The brain is the part a competitor cannot copy by renting the same model: it is the business, written down in a form a machine can work from.

3. Agency inside, authority outside. The agents work freely inside their own computer. Anything that leaves it waits for a person the business has authorized, and who the employee is, who may approve and who may spend are enforced by code the agents cannot argue with. Secrets never enter the model’s context; secret-backed work goes through deterministic adapters that return redacted receipts. The owner approves anything that reaches a customer or touches money, and every action is written down for the owner to check. Payments run on Stripe Connect with AMTECH’s platform fee riding on each charge; invoicing finished work, progress invoices, card and bank payment and automatic reminders are proven end to end in Stripe’s test mode.

4. The gateway. Every model call, picture and second of video any employee makes goes through AMTECH’s own gateway. It meters cost and price per call, sends each call to the best model for the job from any lab, and makes each pricing experiment measurable from day one. When any lab ships a better or cheaper model, every employee gets it the same day. The labs’ race is my supply chain.

5. The factory and the pool. Employees are built from one standard and kept ready, so a person at any door is handed a live employee at the first word, and the employee learns the business from its first job and writes it to its own records. No person is needed to make one; the $1,200 is the managed tier, where I work with the business myself. Next, one business runs several employees that work together, and an employee is trusted to move money inside limits the owner sets, with keys it can never read and a kill switch.

6. The public employee. The same employee can be a web page. On amtechai.com/contractors a visitor describes a job, the employee prices it from that business’s own prices and starts work, answering in about a second once it is open, makes a page the customer can open, and can take payment by card or bank. Every visitor is a free trial, and a page like it can be put up for any trade. One of those runs is published whole on the site as receipt number one: a gutter job priced at $2,912 from the visitor’s own rate, with its two assumptions named.

What the experiments have measured so far.

• Cost to serve, from the gateway’s ledger (7,428 calls to 2026-10-17): typical employee work costs about 5 cents an hour in model calls; the busiest continuous stretch on record cost 46 cents an hour. A whole onboarding interview costs a few cents. A one-week free trial costs about $4 typical, $12 worst case.
• Delivery speed, from the commit record: a moving company’s site, lead handling and back-office integrations were migrated in 4 hours 45 minutes of commits; a service company’s payments and admin took about 25 hours. These are delivery times, not claims of autonomy.
• Pricing, in discovery by design: a setup fee plus usage at about an 85% margin on cost, a flat 50 cents per working hour, and revenue share (8% of a service company’s revenue; 10% of a festival’s money raised) run side by side, each measured by the gateway, so the data and not a guess picks the one owners keep.
• Email, from a 114-message outreach corpus: the thresholds for what reads as a person writing (paragraph count, sentence length, no tracking, no hard wrap) were measured from the corpus and are enforced as gates in code before anything sends.

Where it stands. The labs serve companies by sending engineers in, aimed at mid-sized firms; Meta and xAI launched agents for businesses this fall. AMTECH runs a team of agents for each business as a managed service, on the best model from any lab, and can be paid from what the business earns. For a service business turning over $1 million to $10 million a year, a well-configured team of agents can run essentially everything except the physical work. The 36.2 million small businesses in America hold almost half of its private-sector jobs, and an engineer per company will never reach them; a product can.

ben@amtechai.com · 805-886-9173 · amtechai.com

AMTECH AI LLC · Dalton, PA · prepared October 2026 · companion to the deck and the one page.

The deals, October 2026

AMTECH is a platform for creating, running and managing AI employees, each a team of AI agents with a computer of its own. The platform’s first line of code was written on August 24, 2026. Four deals were signed in October, each on its own terms. The price shape is a setup fee, then usage; two deals add a share of what the business earns.

Clients are named by kind until each gives permission. Before the platform, the founder built websites and systems for paying clients in Pennsylvania, California and Texas; that was freelance work and is not counted as AMTECH revenue. AMTECH has run its own business on one of its employees since August.

What it costs to run an employee, measured

Every model call runs through AMTECH’s own gateway, which meters cost and price per call. Cost is read from that ledger (7,428 calls to 2026-10-17): typical employee work costs about 5 cents an hour in model calls; the busiest continuous stretch on record cost 46 cents an hour. The basis below is an employee working 3 hours a day, 5 days a week, with a few scheduled jobs.

Usage is billed today at about an 85% margin on that cost. A flat price of 50 cents per working hour is in test; at that price the 65 hours above bill $32.50 a month. The price of a fixed level of AI capability has been falling 9 to 900 times a year (Epoch AI), so the cost line falls with every model release. AMTECH is in price discovery on purpose: setup plus usage, a flat hourly price, and revenue share run side by side, and the gateway measures each one.

The round

For the round size, Carta’s Q2 2026 median post-money cap is $12.5M (rounds of $500k to $999k). The $6M cap sits under it on purpose. The founder sets the final terms.

What the $600,000 does

It pays for the computers and the AI the employees run on, the free trials, getting owners to the page, and the company’s running costs, with a reserve. The founder does the meetings and the go-to-market himself with AI agents, so none of it goes to hiring. At the end of eighteen months any owner who finds a page can sign up without him.

Arithmetic on the terms (not a forecast)

A hundred businesses on the managed terms is $120,000 in setup fees alone, before any usage and before either revenue share. The table below is what the signed terms produce at a given number of businesses, so a reader can do the sum; the counts are inputs, not predictions. Managed usage is taken at the typical cost above billed at an 85% margin (about $115 a month); self-serve at the 50-cent test price (about $32.50 a month). Revenue-share deals are left out because their size depends on the business; one such deal on a busy business is worth many times the average.

The cost line uses today’s measured cost, which falls with every model release; the price lines use today’s prices, which are in discovery. Neither is a projection of what the company will do in a year.

Cap table, before the round

AMTECH AI LLC is a Pennsylvania LLC wholly owned by its founder: no prior investors, notes or SAFEs. Any change of company structure a deal needs is handled when that deal needs it.