Perplexity launched something this week that should change how you think about staffing. It's called Perplexity Computer. For $200 a month, you get an AI agent that operates your existing software autonomously — browsing, clicking, filling forms, completing multi-step workflows. Not answering questions. Doing work. On the same tools your team uses every day.
If you read yesterday's piece on the Waymo ratio — one human overseeing 43 autonomous vehicles — this is the economic sequel. That article asked what your practice looks like at a fundamentally different staffing ratio. This one answers what it costs.
Two hundred dollars a month.
The pricing spectrum you can't ignore
Perplexity isn't alone. OpenAI is pricing its agent products at $2,000 to $20,000 per month — and calling them employees, not tools. SoftBank committed $3 billion to OpenAI's agent platform this year. Average enterprise AI spend hit $85,000 a month in 2025, up 36% year over year. The share of companies planning to spend more than $100,000 monthly has more than doubled.
That's not experimentation. That's infrastructure spending. And $200 is the entry point, not the ceiling. A new labor category is forming — digital employees at every price point, for every tier of work. The floor keeps dropping while the capability keeps rising. This week alone, Claude 3.5 Haiku got significantly smarter at half its previous cost. GPT-4.5 launched with a focus on nuance and judgment. Grok 3 shipped with 10x the compute of its predecessor. Every model improvement makes the $200 agent more capable without raising the price.
The readiness problem is bigger than the price
Here's what concerns me. The price is now so low that every practice owner reading this could buy one tomorrow. And most of them would get nothing from it.
Because a $200 digital employee doesn't slot into your current org chart. It doesn't know your chart of accounts. It doesn't understand your client naming conventions or your review process or when to escalate a question to a senior. It doesn't know what "done" looks like at your firm.
The firms that buy an agent and bolt it onto their existing workflow — same roles, same processes, same handoffs, just cheaper — will capture almost nothing. We've talked about this pattern before. Bolt-on versus rebuild. The price point makes the bolt-on temptation worse because $200 feels disposable. Worth a shot. No risk. Except the risk isn't the money. It's the conclusion you'll draw when it doesn't work — that AI agents aren't ready — when the real problem is that your practice isn't ready for them.
The demand signal was already deafening
Moltbot — the open-source AI agent that became the fastest-growing project in GitHub history — proved the appetite. Over 100,000 people gave an AI autonomous access to their digital lives within weeks. Perplexity Computer is the polished, commercial, managed version of that same demand. The wild experiment just became a product you can buy.
But the lesson from Moltbot still applies. The difference between the agent that saved someone $4,000 negotiating a car purchase and the one that carpet-bombed a contact list with 500 messages was one thing — the quality of the specification. The context engineering. The employee handbook you write for the AI before you let it touch anything.
What your team actually needs to learn
This is the part most practice owners skip. Your existing staff aren't being replaced by the $200 employee. They're being asked to manage it. That's a fundamentally different skill set, and almost nobody has been trained for it.
Reviewing AI output instead of producing it. Designing workflows instead of executing them. Knowing when to trust the agent's work and when to intervene. Setting the guardrails that keep a digital employee inside the boundaries of your firm's standards. Catching what the agent misses — and knowing what it's likely to miss.
That's not intuitive. Your bookkeeper knows how to do a bank rec. They don't know how to supervise an AI doing 30 bank recs simultaneously and catch the three that need human judgment. Your senior knows how to review a staff accountant's work. They don't know how to review an agent's output at 10x the volume with a completely different error profile.
The human role doesn't shrink. It transforms. And the transformation requires training your team has never had.
Agent orchestration is the unlock
Our AI Black Belt program builds through five layers — from foundational AI literacy through prompt architecture, context engineering, and workflow automation. The fifth layer is agent orchestration. It's the capability that makes everything else in this article possible.
Agent orchestration isn't prompting a chatbot. It's designing multi-step workflows where AI handles execution and humans handle judgment. Setting guardrails. Managing handoffs. Building the specifications — the context engineering we've been talking about all year — that turn a $200 generic agent into a $200 agent that operates inside the specific reality of your practice.
Without that layer, you're buying a tool. With it, you're hiring an employee.
The math underneath your staffing decisions just changed
Yesterday we talked about what the ratio looks like. Today we're talking about what it costs. A bookkeeper runs $5,000 to $8,000 a month fully loaded. A digital agent handling a meaningful share of execution work costs $200 to $2,000. The ratio redesign we described yesterday isn't just structurally compelling — it's economically unavoidable.
Your fee structure is either a massive margin opportunity or a competitive vulnerability right now. If your delivery cost drops 40 to 60% and you're still pricing like it hasn't, you're choosing to leave margin on the table. If your competitor figures this out first, they reprice to win your clients while maintaining profitability you can't match.
The firms that train their teams now set the terms. The firms that wait get repriced by competitors who did.
If you want your team ready for this, the AI Black Belt program is built for exactly this moment. Five layers. From foundations through agent orchestration. The full capability stack your team needs to work alongside digital employees — not just use AI tools. Visit theaiaccountant.ai to get started.
Would you hire a $200/month employee? The better question is whether your practice is ready to manage one.