Why we're building Sam

In February 2026 a company called Viktor put an AI agent inside Slack, gave it a computer and access to a few thousand tools, and let teams ask it to do things. Ten weeks later it reported a $15 million revenue run rate and raised $75 million. Businesses had been waiting for software that finishes the task instead of handing it back.

We think that was the first draft of something every company will have. We also think the first draft has problems that stop it at pilot scale, and that those problems are the product.

What the first generation got right

An AI employee has to live where the team already is, have its own computer, connect to the company’s systems through real credentials, and come back with finished work: a PDF, a changed record, a pull request, a dashboard. It has to run on a schedule without being asked twice. Viktor’s customers describe replacing a social media tool, a part-time assistant and a handful of subscriptions with one bot. That is real.

What stops a business from going further

Read the reviews and the same five things come up.

The bill is a surprise. Token pricing means you pay for the agent’s verbosity and for its retries. You cannot know what a task will cost before it runs. Teams exhaust a month of credits in a day and go back to doing the work themselves.

Nobody can explain what it did. There is no record a manager or an auditor can read. “Why did you change that?” has no answer.

One bot, everyone’s credentials. A single shared identity per workspace with every integration pooled is fine for four people and unacceptable for forty. Per-user permissions are on roadmaps, not in products.

It only lives in Slack. Most businesses on earth don’t. The accountant, the clinic, the property manager and the plumber run on email, WhatsApp and the phone.

Compliance is a hero line, not a date. “SOC 2 compliant” above the fold and “Type II in progress” in the FAQ. Procurement notices. And since 2 August 2026 the EU AI Act requires that people know when they are talking to an AI, which an agent that sends email and makes calls has to get right every time.

What Sam is

Sam is the AI employee you can hire, trust and audit.

• You hire it: a role description, a manager, a budget, its own email address and phone number, a probation report at thirty days, a weekly 1:1, a clean offboarding.
• It shows the plan and the price before it runs and absorbs the cost of its own mistakes. You set hard budgets.
• It keeps a work log of everything: every read, every write, every model call and what it cost. Replay it, diff it, undo it, export it.
• It works where you work: Slack, Teams, Google Chat, Discord, email, WhatsApp, SMS and the phone, and a task can move between them.
• It acts as you, with your permissions, or as itself with its own, and it keeps what it learns inside the boundaries you set.
• It is compliance-native: dated commitments on the trust center, EU residency, disclosure to everyone outside your company that cannot be switched off, and packs for the EU, healthcare, finance and HR.
• It deploys anywhere with any model, including your own cloud and self-hosted.
• It never says “I can’t”: integration, MCP, browser, desktop, phone call, or a vetted person for the physical world, each priced on the plan card.

What happens next

We are starting with ten design partners in a private alpha. If your team wants an AI employee and has been told no by your own security review, or if your team has never opened Slack, we would especially like to talk. The waitlist is on the front page.