Viktor: AI Coworker That Lives in Slack — Fryderyk Wiatrowski
TL;DR
Fryderyk Wiatrowski presents Victor, an AI employee that lives natively in Slack to automate complex cross-functional tasks by leveraging shared company context and 3,000+ tool integrations, evolving from early browser-based agents to solve the unique memory and permission challenges of multi-user enterprise environments.
🚀 Product Evolution & Architecture 2 insights
From unreliable browsers to Slack-native
Early browser agent JCAI achieved only 60% reliability on 3-5 step tasks via DOM snapshots in 2023, prompting evolution into email agent Jace and the February 2024 launch of Victor as a company-wide AI employee with immediate product-market fit.
Shared integration model
Unlike personal agents requiring individual connections, Victor inherits permissions when one employee connects an integration, providing universal PhD-level context across codebase, analytics, and marketing tools for the entire team.
🏗️ Technical Challenges in Slack 2 insights
Multi-user memory isolation
Solved the 'cluttered memory' problem that scales exponentially with 100+ users by architecting strict context isolation between Slack channels and DMs based on team hierarchy to prevent executive data from leaking to engineering channels.
Non-linear interaction handling
Engineered systems to interpret Slack's complex social signals including edited/deleted messages, abandoned threads, and emoji reactions as task modifications or cancellations, converting these into linear agent context.
🧠 UX Psychology & Deployment 3 insights
Latency arbitrage
Selected Slack because users tolerate 10-minute response times for complex tasks from 'coworkers' versus the immediate gratification expected in web apps, making powerful agentic workflows feel natural rather than frustrating.
Personality drives adoption
Users rejected GPT-4 in favor of Claude Opus during A/B testing due to Opus's preferred 'sassy' and human-like tone, demonstrating that emotional resonance significantly impacts workplace AI adoption more than raw capability.
Graduated proactivity
Recommends restricting proactive suggestions to initial power users before company-wide rollout to prevent security team backlash from sudden autonomous DMs and thread participation.
💼 Integration Philosophy 1 insight
Treat as a hire, not a tool
Implemented scoping controls allowing integrations to be personal or shared after users mistakenly granted Victor access to personal Gmail, emphasizing that AI employees require appropriate permission boundaries just like human hires.
Bottom Line
Successful AI coworkers require shared company context, distinct personality, and graduated deployment in Slack to leverage social latency tolerance while solving multi-user memory and permission architectures.
More from AI Engineer
View all
The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks
Sandipan Bhaumik from Databricks presents a battle-tested five-pillar framework for deploying enterprise AI agents, arguing that starting with model selection leads to inevitable production failures while proper evaluation, observability, and data governance determine success at scale.
Sovereign Escape Velocity: Ownership w Open Models — Gus Martins, & Ian Ballantyne, Google DeepMind
Google DeepMind's Gus Martins and Ian Ballantyne introduce Gemma 4, a family of open models (2B to 31B parameters) that deliver frontier-level intelligence with disproportionate efficiency, enabling sovereign AI ownership through local deployment, Apache 2.0 licensing, and on-device capabilities.
LLM Observability, Evaluation, Experimentation Platform — Dat Ngo, Arize
Dat Ngo from Arize AI explains how modern AI systems require reimagined observability and evaluation patterns built on OpenTelemetry to manage non-deterministic agents, emphasizing that the future of AI engineering lies in automated experimentation flywheels that eliminate manual dashboard work.
Text Diffusion — Brendon Dillon, Google DeepMind
Google DeepMind researcher Brendon Dillon explains text diffusion as a parallel alternative to autoregressive language models that iteratively denoises random tokens rather than generating sequentially, offering significantly lower latency and unique capabilities like self-correction and adaptive computation, though currently limited by high serving costs for large batches.