Kylon vs. Viktor: AI coworker in Slack vs. AI-native workspace

Viktor launched in March 2026 and hit 12,000 workspaces in under three months. A $75 million Series A from Accel followed in May. The pitch is straightforward: add one AI employee to Slack or Microsoft Teams, connect your tools, and let it do real work — dashboards, reports, campaigns, code commits, research, browser automation.

It's a good pitch because it's a real product. Viktor doesn't summarize documents or generate paragraphs. It logs into your Stripe, pulls live data, cross-references with HubSpot, and delivers a board-ready PDF. It writes code, opens pull requests, and deploys internal tools. It runs on a schedule, monitors anomalies, and proactively suggests automations. Former Meta engineers Fryderyk Wiatrowski and Peter Albert built it, and the execution quality shows.

We think Viktor is doing something right. We also think there's an architectural ceiling to the approach.

What Viktor does well

Zero-friction deployment. Install a Slack app, connect your tools through OAuth, and start messaging. No infrastructure, no configuration, no developer required. For teams that live in Slack, the time-to-value is remarkably short.

Genuine output, not chat. Viktor delivers PDFs, spreadsheets, deployed web apps, code commits, and structured reports. It doesn't just talk about work — it produces artifacts. This is a meaningful distinction from most AI assistants.

Breadth of integrations. 3,200+ tools connected through managed connectors. Stripe, HubSpot, Google Ads, Meta Ads, PostHog, Linear, Notion, GitHub, Google Calendar — the coverage is impressive and saves teams from building custom connections.

Proactive automation. Viktor's "heartbeat" system observes how teams work and suggests tailored automations. Morning revenue briefings, weekly ad audits, anomaly alerts. Approve once and it runs on autopilot. This is closer to how a real employee operates — noticing patterns and acting on them.

Browser automation. Viktor operates a real browser — filling forms, navigating workflows, scraping data. For tasks that don't have APIs, this is a practical workaround.

SOC 2 certified. Type 1 complete, Type 2 in progress. For enterprise buyers, this matters.

Where the architecture diverges

Viktor and Kylon share the belief that AI should do real work, not just answer questions. The difference is where the AI lives — and what that implies for data, coordination, and scale.

One employee vs. a team

Viktor is designed as one AI employee. One persona, one identity, one set of capabilities shared across the whole organization. Everyone in the Slack workspace talks to the same Viktor.

Kylon is designed for multiple specialized agents working alongside humans. A marketing agent that knows your campaign history. A data agent that understands your metrics. An ops agent that manages workflows. Each has its own identity, memory, skills, and permissions. They coordinate with each other and with humans in the same channels.

The difference matters when work gets complex. A single generalist employee — no matter how capable — has limits. A team of specialists that share context and hand off tasks scales differently. When the marketing agent needs data, it asks the data agent. When the ops agent completes a workflow, it notifies the relevant humans. The coordination is native, not bolted on.

Guest in Slack vs. native workspace

Viktor lives inside Slack (and now Teams). That's a strength for adoption — but it means Viktor inherits the host platform's constraints.

Data lives elsewhere. Viktor can pull data from tools, but it doesn't own a data layer. Insights, reports, and analyses are delivered as attachments or messages. There's no structured database where agent-generated knowledge compounds over time.

Kylon has a native data layer — database apps that agents read, write, and query directly. When an agent analyzes your revenue trends, the results live in a table that other agents and humans can reference, filter, and build on. Data compounds inside the workspace.

Threading and context are Slack's. Viktor's conversations follow Slack's chronological message model. When agent output scales — dozens of reports, dashboards, and analyses per day — it mixes with human conversation in a flat feed. Finding what Viktor produced last Tuesday means scrolling through a channel.

Kylon's information architecture is built for agent-scale output. Channels, threads, tables, apps, and files are structured for both human and agent work. An agent's output can be a table row, a deployed page, a workflow result, or a structured card — not just a message in a feed.

UI is Slack's. Viktor's output format is constrained by Slack's Block Kit — a fixed set of UI components designed for human messaging, not for agent output. Charts, interactive tables, conditional buttons, user-specific views — these are limited or unavailable.

Kylon agents render output in whatever format communicates most efficiently. Interactive cards with drill-down sections. Real tables with sortable columns. Stats dashboards. Approval workflows with user-specific buttons. Different views for different roles in the same message.

Cross-platform vs. Slack-bound

Viktor requires Slack or Teams. If your team works across devices — desktop, phone, tablet — you need Slack installed and active.

Kylon runs natively on web, macOS, iOS, and Android. The workspace meets your team wherever they are, with the same agent capabilities across every platform.

Permissions and accountability

Viktor's integrations are workspace-level — connected once, available to everyone who can message Viktor in that channel. This is simple to set up but means the permission model is coarse. If Viktor has write access to GitHub, anyone can ask it to open PRs.

Kylon agents have individual identities and per-agent permissions. Each agent has defined capabilities, specific tool access, and an audit trail. You know which agent did what, when, with what context, and under whose authorization. Permissions are granular — an agent's access to a connection can be scoped by user, by channel, or by the specific action requested.

Workflows and scheduling

Viktor handles scheduled tasks well — daily standups, weekly audits, monthly reviews. But the scheduling is within Viktor's own system.

Kylon has a composable workflow runtime — manual triggers, cron schedules, webhooks, row-change triggers, and multi-step pipelines. Workflows can invoke agents, chain operations, and run code. It's a programmable automation layer, not just a task scheduler.

The broader question: tool or workspace?

This is the same architectural question that Claude Tag raises. When AI reaches a certain level of capability, is it better to embed it inside an existing tool — or to build a purpose-designed workspace around it?

Viktor makes the "embed" bet brilliantly. The Slack integration is seamless, the output quality is high, and the breadth of tool connections is impressive. For teams that are all-in on Slack and want a single capable AI employee, Viktor delivers genuine value.

Kylon makes the "workspace" bet. When you control the entire environment — data, UI, permissions, workflows, platform — you can build things that a guest in someone else's house can't. Multi-agent coordination. Native databases. Cross-platform access. Granular permissions. Rich, agent-optimized output formats.

Both bets have merit. The question is what your team needs.

When each makes sense

Choose Viktor when your team lives in Slack and wants a single, capable AI employee that connects to everything. If you need dashboards, reports, and automations delivered into your existing channels with zero deployment friction — Viktor is polished and production-ready. The $50/month Team plan (20,000 credits) makes it accessible for lighter usage, though complex workflows can consume credits quickly, and the output quality is genuinely impressive.

Choose Kylon when you need more than one AI role, or when you need the AI to own data and workflows, not just produce artifacts. If you want specialized agents with individual permissions, native databases that compound organizational knowledge, multi-agent coordination, composable workflows, and a cross-platform workspace that runs on Mac, iPhone, and Android — Kylon is the purpose-built environment for human-agent teams.

Two takes on the same future

Viktor proved that teams will adopt AI that does real work — not just AI that talks about work. 12,000 workspaces in three months is strong signal. The market clearly wants AI employees.

The question is whether the future is one AI employee inside your existing tools, or a team of AI agents inside a workspace designed for them. We think it's the latter. But Viktor's success validates the core thesis both companies share: AI should do the work, not just discuss it.