With lower costs through models like Deepseek’s R-1 and better reasoning, we’re going to see more and more companies make use of them at the application layer through agents. Over the last year, we’ve seen a rapid evolution from AI copilots—tools that simply guide or augment a human—to full-fledged AI agents that can do more tasks autonomously. One common way we’re seeing them commercialized within organizations in the form of AI teammates/ colleagues/ coworkers. These teammates often mirror familiar roles (like an SDR, support rep, or test engineer), and are able to handle a subset of tasks that humans in those roles do today.
As more AI colleagues pop up across roles, we’re witnessing the emergence of the “agentic workforce”—specialized AI agents that slot into specific roles within a functional area or vertical, promising to give organizations more leverage in that area.
The agentic workforce across roles
What’s really striking is the sheer pace at which AI agent based startups have sprung up in nearly every function—from sales to HR to marketing—packaging themselves as AI teammates that promise to offload real workloads. Below is a market map of the workforce across key functional roles that illustrates just how quickly these have popped up everywhere:
I’ve kept this focused on functional roles (an overview of vertical roles will come in the future) and only included startups that are actually agentic and to some extent brand / market themself as an AI coworker/teammate, and been strict to not include AI services businesses in this (which as I’ve written about in the past are a different archetype of AI startup). You’ll notice that there are many missing roles or departments, such as design, product, or AEs where I don’t believe there is an agentic coworker yet.
If you notice any company missing or think there’s a role/functional that deserves its own row, please let me know by commenting or sending me an email at tanay@wing.vc — I’d love to update the market map and keep it as comprehensive as possible.
Characteristics of agentic workers
Let’s discuss some of the key traits common to these new AI colleagues:
- Skeumorphic roles: These agents are modeled after real positions we already recognize—like an AI SDR or AI Support Rep—which makes their capabilities easier for prospective buyers to digest.
- Work-focused, not software-focused: The core offering isn’t “software features” to help humans do work but rather AI that can do work directly itself. Instead of buying software, you’re effectively hiring an AI to do parts of the job.
- Task-based capabilities: Even though marketed as AI versions of an entire role, most start with just a few tasks in their wheelhouse that AI is good at and as they earn the trust of their human managers (and their capabilities improve with model and other improvement), they can expand their responsibilities. Today, most are far off from the performance of humans in that role, but they can supplement or provide more leverage, and do certain tasks with more consistency, speed or accuracy than humans.
- Human management: There’s typically a real person in the loop at the customer end who is “managing” and directing the AI agent, assigning or reviewing tasks or intervening as needed. For example, IC engineers at a customer will onboard, chat with and direct Devin, Cognition’s AI engineer. To be clear, this is different from startups where the human in the loop is entirely on the end of the startup itself (where the startup is selling fully managed services). Those are AI services business and distinct from these agentic coworker businesses, at least in the medium term.
- Labor budgets, not software budgets: Many of these startups are targeting labor budgets rather than software spend. This helps in a few ways: One, labor budgets are at least an order of magnitude larger than software spend is almost every area which means larger TAMs and potential ACVs. Two, these companies don’t have to displace or rip out existing software, and in many cases, like a human would, can work and operate within the existing software stack, reading/writing to it as needed.
- Usage/Outcome-based pricing: Rather than per-seat or enterprise-wide licenses, these agentic workers often charge by some form of usage (e.g., data processed, tasks done) or outcome (e.g. ticket successfully resolved). This helps them better reflect the value they provide, but does mark a shift from the seat-based pricing that SaaS has been built on over the last few
Needless to say, we at Wing are very excited about the agentic workforce opportunity — if you’re working on AI colleagues across any vertical or functional role, get in touch on X or email me at tanay@wing.vc.
