3 years ago, the best compliment you could pay a piece of software was that it made your team faster.
Today, the best software IS your team.
That sentence would have sounded like marketing in 2023. It doesn’t anymore, and this shift is the biggest change to knowledge work I have seen in my career.
If you asked me how I’d describe the last 3 years, here’s how I would think about them -
Phase 1: software as assistant.
When AI first started becoming mainstream across our tools, it lived inside the workflow. It autocompleted your email, summarized your meeting, suggested the next line of code. It was helpful and sometimes even felt like magic. But the shape of work didn’t change much. You still had to open the tool, give approval at each step, and own the outcome. The software had to wait for you.
The way we even measured it was different: seats, licenses, adoption. The assumption baked into all of them was that a person sits in the chair and the software sits in the background.
Phase 2: software as workforce.
Somewhere in the last 18 months, the question changed. We stopped asking “how does this tool help my team do the work?” and started asking “which of this work can the tool just do by itself?”
That is a way bigger change than it seems on the surface. A workflow waits for a human to push work through each step. An agent takes the goal and runs through the steps itself - gathering context, making judgment calls at each fork in the road, and coming back when its done or when its genuinely stuck. The human goes from being an operator to a manager of a fleet of agents.
I’ve seen this from both sides. In my time at Plaid, I watched great post-sales teams run tight workflows: a sales engineers week was spent wrapping up a successful pilot and doing the post-sales handoff, an account manager’s week was spent digging through usage data, writing follow-ups, logging notes in the CRM, and using all of it to identify expansion opportunities and mitigate churn. The software organized that work - in nice clean looking tables, dashboards, tabs. Ultimately a person still had to execute all of it.
At Quivly we’ve built agents that run the post-sales motion end to end - onboarding, adoption, expansion, renewals - in the background, across every account at once. The people doing the work are still there. The nature of their job has changed: less assembly, more judgment. Fewer hours preparing for the customer conversation, more hours in front of the customers, having the conversations that actually matter.
Once you see this in one function, you see it everywhere. Engineering teams managing agents that write and review code. Sales teams where agents run research and first touches. Support orgs where agents hold the front line. Every knowledge function at the most forward thinking companies is already split into two buckets: work agents do, and work that still needs a person.
Which brings me to the part that I think a lot of companies are getting wrong.
They are buying agents and managing them like software.
An agent is not a tool you install. It’s much closer to a new hire on your team. It needs onboarding - your context, your data, your definition of what good looks like. It needs review, because it will occasionally be confidently wrong in ways a dashboard could never be because it didn’t do the work for you. And it needs a manager who decides what it’s allowed to do alone and what escalates to a human.
Every leader I talk to is facing the same choice:
1) Treat agents as tools - another line in the software budget, measured in seats.
2) Treat agents as a workforce - hired, onboarded, reviewed, and managed.
Option 1 gets you slightly faster workflows.
Option 2 changes how your company operates.
And I’ll be honest, we’re all still figuring out what option 2 really means. Agents fail differently than software. When software breaks you’re not able to finish some task but when agents mess up the task has often been finished, just poorly. Managing a workforce that’s part human and part agent is a skill we didn’t need to learn up until 3 years ago. So it’s no surprise we haven’t fully figured it out yet.
The last era of software was about giving workers better tools. This one is about giving companies more workers.
