FirstMark Partner Matt Turck: Talking to People in 2016, Talking to Zoom in 2021, Talking to AI Agents in 2026
FirstMark partner and author of the MAD industry map, Matt Turck, summarizes the evolution of work in three lines: talking to people in 2016, talking to Zoom in 2021, and talking to agents in 2026.
This is not a product launch but a compression of the transition in office interfaces. 2016 corresponds to office and travel meetings; 2021 corresponds to video conferencing becoming the default collaboration layer post-pandemic; 2026 corresponds to delegating tasks to executable software agents instead of holding another meeting. Turck has long tracked data and machine learning applications, including customer service agents, enterprise intelligence, and automation companies, and hosts Data Driven NYC and the MAD podcast.
The same week's product events provide footnotes to these three lines: HR platforms are outsourcing internal agents and claiming to replace hundreds of full-time equivalents with agents; coding editors are integrating charts and multi-agent windows into conversations; in-car voice systems are connecting documents and external coding agents into the steering wheel. Video conferencing has not disappeared, but an additional layer has emerged where "you can leave after speaking."
The three lines omit the intermediate state: chatbots from 2023 to 2025 are still stuck in drafts and Q&A, while the narrative for agents in 2026 emphasizes learning processes through screen recording, integrating with enterprise systems, and continuing to operate in the cloud. Investors use a ten-year scale to write the interface change as irreversible, rather than focusing on the market share of a specific app.
In market mechanisms, what is bought is early funds that treat "agents equal to new interfaces" as a configuration principle, while what is sold is the next layer of entry after meeting software. Funding is shifting from collaborative seats in enterprise software budgets to automation billed by task or agent; beneficiaries are agent vendors that can connect to finance, HR, and code repositories, while those under pressure are tools that only sell meeting room minutes and cannot finish the conversation. Event-driven is a definition, not a financial report.
Talking to people consumes coordination, talking to video consumes schedules, and talking to agents consumes permissions and audits. Each time the interface changes, the billing unit shifts from headcount to completed work packages.
Source: Public Information
ABAB AI Insight
Turck uses an annual industry map to queue data and AI companies, and the three lines of evolution serve as a slogan for the latest box on the map. Having seen every round of enterprise software sell "new interfaces" as platforms, he transitioned from search startups to Oracle and then to FirstMark. SaaS sales in 2016 still revolved around meeting rooms, while in 2021, Zoom moved meeting rooms into calendars, and by 2026, agents are expected to consume part of the meetings scheduled on the calendar. His portfolio includes both customer service agents and automation operations, with speech and holdings aligned: the next enterprise software check will be written to systems that can turn speech into actionable results.
The capital path is a restructuring of collaboration budgets. Seat fees in the Zoom era are billed per person per month; in the agent era, they will be billed by automated hours, by cases, or by completed workflows. Enterprises will first record dirty processes for agents, then reduce coordination meetings, revealing the growth ceiling for meeting software. Motivations will be reflected in valuation language: per capita output, full-time equivalents, and 90-day incremental revenue are all financial translations of "talking to agents." Whoever defines this set of metrics first will occupy the central application layer on the next MAD map.
Analogies can be drawn from the transition from telephone to email, email to Slack, and Slack to video. Each jump retains the previous layer, but the default action shifts. The industry phase is moving enterprise software from "connecting people" to "removing tasks from people." If Salesforce and ServiceNow only handle work orders for people, they will cede the execution layer to screen recording agents; if Zoom only focuses on faces, it will cede post-meeting actions to agents capable of operating computers.
The structural judgment is one of technological substitution. The mechanism is that once communication media can invoke systems, coordination itself becomes an optional middleware. Substitution occurs not because people dislike their colleagues, but because the marginal output of meetings packed on high-DPI calendars has been surpassed by the backend operations of agents. The remaining conversations with people will become more expensive, shorter, and more like exceptions.
ABAB News · Cognitive Laws
Each time the work interface changes, the billing unit changes.
Software that can finish conversations will consume the next meeting.
In ten years, the three lines have paved the way for the next line.