Forget “Would You Hire Them Again.” The Real Test Now Is Whether You’d Replace Them With an Agent.

For our entire careers in tech, the standard management test has been: “Given what you know now, would you hire this person again?”

It’s a decent HR framework. It forces honesty. If the answer is no, you have your decision.

But in 2026, the question is incomplete. The real test now is: “Would you replace them with an agent?”

That’s a different question. And it’s the one most managers aren’t asking yet.

Why the Old Test Doesn’t Work Anymore

“Would you hire them again” assumes the alternative is another human. It’s a comparative test between two humans doing the same job. In a world where a competent agent costs $200/month and can do 60-80% of most knowledge work, the comparison isn’t human vs human anymore. It’s human vs agent. Or more accurately, human + agent vs agent alone.

If the honest answer to “would you replace them with an agent” is yes, you have your decision. If the answer is no, you need to articulate exactly why. What’s the 20% of their work an agent can’t do? Is it worth a full-time salary plus benefits plus management overhead plus equity?

Most managers have never been forced to think this way. They think about headcount as a given and productivity as the variable. The new model flips it. Agents are the given. Humans are the variable.

The Three Questions Underneath the Test

When you actually run the replacement test, you’re asking three questions:

1. What is the job?

Most job descriptions are a list of tasks, not an outcome. An SDR’s job is not “send 100 emails a day.” It’s “generate qualified pipeline.” A marketing manager’s job is not “run campaigns.” It’s “drive demand.” Once you articulate the outcome, you can ask what mix of humans and agents gets you there most efficiently.

2. What can an agent actually do?

Not what a demo shows. What can an agent do in production, reliably, at scale, with minimal oversight, today. For SDR work, that’s 80-90% of the job. For VP of Marketing, it’s maybe 40% today but climbing. For CFO, maybe 30%. For CEO, close to zero.

The skill here is being precise about capability. Too many execs either dismiss agents because they tried one that was bad, or oversell them because they saw one demo that was good. The real answer requires actually running them in production for 3-6 months.

3. What’s the remaining human work worth?

This is the hardest question. If an agent can do 70% of a role, is the remaining 30% worth a full-time human? Maybe. Maybe not. Sometimes that 30% is the judgment that matters most and absolutely requires a human. Sometimes that 30% is “being the face of the function” and could be consolidated under a more senior person running multiple functions with agent support.

The answer varies by role. But you have to ask.

What This Looks Like in Practice

At SaaStr, we run the company with 3 humans and 20+ agents. Revenue went from -19% to +47% YoY after the shift. That’s not a rounding error. That’s a structural change in what the company is capable of.

What stayed human: strategy, judgment calls, relationships that require trust at the top, and the things where the human voice matters more than the task.

What got replaced or augmented by agents: SDR work (Artisan, Qualified, Monaco, others), win-back campaigns (Agentforce), customer success workflows (our own AI VP of Customer Success QBee), content production support, sponsor coordination, analytics on portfolio companies. Not all of it perfect. But all of it cheaper, faster, and measurable.

The lesson isn’t “replace everyone with agents.” The lesson is that the composition of the team is now a variable, not a constant. Every open req is a chance to ask: do we need a human here, or do we need an agent plus a supervisor?

The Skill That Matters in 18 Months

Here’s the part most people are missing. In 18 months, everyone will know how to build a good agent. Prompt engineering was a hot job title for 12 months and then became irrelevant. Agent building is on the same trajectory. The tools get easier. The models get smarter. The agent frameworks commoditize.

The skill that matters isn’t building the agent. It’s deciding which humans to replace with one.

That’s a management skill, not an engineering skill. It requires judgment about capability, honesty about performance, and a willingness to have difficult conversations. Most managers are not trained to think this way yet. The ones who learn fast will run leaner, faster companies. The ones who don’t will be the ones getting replaced themselves.

How to Actually Run the Test

Every quarter, or at minimum every time you’re about to hire for a role:

• Write down the outcome the role is supposed to produce, not the tasks.
• List what an agent could do today toward that outcome. Be specific. Not “helps with marketing.” What exactly could it do this week, in production.
• Estimate what’s left for the human. Both the tasks and the judgment calls.
• Ask whether that remainder is worth the fully-loaded cost of a human hire. Salary + benefits + equity + management overhead.
• If yes, hire the human and pair them with the agent. If no, don’t open the req. Build the agent instead.

For existing team members, you can run a gentler version. Not “should I fire this person.” But “if I were building this team from scratch today, would this role exist as a human role, or would it be an agent plus a fractional supervisor?”

That’s an uncomfortable question. It’s also the right question.

This Is Not a Layoffs Story. This Is a Choice Story.

The most important thing to get right about this shift: it’s not about cost-cutting. It’s not about layoffs. It’s not about efficiency as a defensive response to a bad market.

It’s about choice.

We will choose to replace many roles with agents because it’s easier, more efficient, and more scalable. Not because we have to. Because we want to. Because running a company with 3 humans and 20 agents is a better way to operate than running it with 30 humans. Faster decisions. No politics. No Slack noise. No meetings about meetings. Just output.

And the honest truth: great AI agents avoid so much pain.

No quitting. No two-week notices right before your biggest launch. No human drama. No passive-aggressive Slack messages. No resentment over working a weekend. No resentment over doing something not explicitly in the job spec. No promotion drama. No “I deserve a bigger title” conversations. No deadline drama. No missed deadlines because someone’s partner is mad at them. No HR issues. No performance improvement plans. No awkward exits. No backfills. No re-hiring.

Every founder reading this knows exactly what I mean. You’ve lived through all of it. Probably this week. The agent doesn’t do any of it. The agent shows up, does the work, and doesn’t need a 1:1 to process its feelings about the quarterly roadmap.

That’s not a small thing. Human drama is one of the most expensive line items in any company, even though nobody puts it on the P&L. Agents remove it almost entirely from the parts of the business where they run. That alone is worth the switch for many founders, even before you get to speed, cost, and scalability.

This is the part that’s hard to internalize if you came up in traditional SaaS. The old mental model was: we grow, we hire, we add headcount, we scale. More people equals more capability. The new mental model is: we grow, we add agents, we maybe add a few senior humans to supervise them. More agents equals more capability. Humans become the expensive, high-judgment layer, not the default execution layer.

Founders who make this shift aren’t doing it reluctantly. They’re doing it because they’ve tried both models and the agent-first model is genuinely better. Less friction. Faster iteration. No vacation coverage to worry about. No performance management drama. No backfills. The agents just keep running.

The companies that treat this as a layoff story will get it wrong. They’ll cut the wrong people, fa