AI Agents Unbundling Office Tasks

Diving deeper into

Filip Kozera, CEO of Wordware, on the rise of vibe doing

Interview
When you start seeing how many white-collar jobs these systems are eliminating, it's scary.
Analyzed 4 sources

The real disruption is not one giant wave of unemployment, but thousands of small office tasks getting unbundled from the people who used to do them. In this market, AI agents first eat repetitive work like inbox triage, meeting prep, scheduling, document updates, and draft writing. Sauna is built around that exact layer of work, which is why the job impact shows up fastest in coordinator, assistant, recruiter, and analyst workflows rather than in fully autonomous replacement of entire roles.

  • The product map makes the labor impact concrete. Sauna is aimed at chief-of-staff work, project management, and artifact creation, which means work that used to require humans to monitor email, chase people, reconcile notes, and turn scattered inputs into polished outputs.
  • Early winners in the category are narrow tools that automate one painful workflow at a time. Fyxer started with inbox labeling and reply drafting for brokers, consultants, and fractional executives, then expanded into scheduling and meetings. That shows how white-collar automation is landing first as software that removes slices of admin labor, not whole departments overnight.
  • The competitive pressure comes from both startups and model platforms. Specialist products like Lindy can win by doing one job well, while Anthropic, OpenAI, and Perplexity can bundle similar agents into tools people already use. That pushes the market toward fewer humans doing more coordination with AI handling the routine throughput.

From here, the category moves from assistant features to always-on work systems that watch inboxes, calendars, chats, and files in the background. As these systems get better at retaining context and acting across tools, the next step is not just faster knowledge work, but leaner org charts with fewer support layers between decision makers and execution.