AOaaS vs Vertical SaaS: Why the $13T Labor Budget Beats the $1.5T IT Budget

By Abhishek Barua, Astra Space AI

Every analyst report I've read starts the same way: "The AI software market is $1.5T."

They're not wrong. But they're looking at the wrong number.

The actual addressable market for what we're building — Agentic Operating Systems — isn't $1.5T in IT spend. It's $13T in labor spend. And there's a massive gap between where enterprise AI companies are building and where the real value lives.

Gartner, Forrester, IDC — they all measure AI adoption through the lens of IT infrastructure and software. Makes sense; that's been the TAM for the last 20 years. CRM software: measure it by IT budget. ERP: measure it by IT budget. Data warehouse: measure it by IT budget.

But Agentic Operating Systems aren't tools for IT. They're replacements for labor.

When you deploy an AOaaS — a platform that lets you spin up AI agents to handle knowledge work — you're not optimizing your IT spend. You're replacing headcount. You're multiplying the output of your existing team. You're letting your $300K senior engineer handle 10x more problems without hiring 10 people.

And the budget for that lives in a completely different ledger: the labor budget. $13T globally.

That's the Bessemer vertical-AI thesis in a nutshell: the companies that win in the next 10 years won't be the ones optimizing IT efficiency. They'll be the ones that unlock labor multiplication.

Why Vertical SaaS Misses the Mark

The last wave of successful enterprise software — Vertical SaaS — was built for a different world.

Vertical SaaS companies like Veeva, Procore, and Figma dominate their categories because they solved a vertical problem better than horizontal players. Veeva dominates pharma CRM because it understands pharma workflows. Figma dominates design collaboration because it's built for design teams.

That thesis still works. But it's reaching its ceiling.

Here's why: Vertical SaaS optimizes the way people work. It doesn't replace people.

A pharma sales rep using Veeva is still a pharma sales rep. They're just more efficient. Figma made designers 30% faster, but you still need the designer.

Agents are different. They replace the function, not just optimize the tool.

This is why the TAM looks different. With Vertical SaaS, you measure adoption in seats (e.g., "How many sales reps use our tool?"). With AOaaS, you measure labor multiplication: "How many reps' worth of output can one engineer produce?"

The ROI math is night-and-day different. A $10K/year SaaS tool that makes you 15% faster? Fine. An AOaaS platform that lets you handle 3x the workload with 1/3 the headcount? That's a $1M+ decision. That's a labor budget decision, not an IT budget decision.

What AOaaS Actually Is (And What It Isn't)

Here's where a lot of companies get confused.

AOaaS is not "AI tools for your team." It's not "ChatGPT for Salesforce." It's not a plug-in.

An Agentic Operating System is an autonomous platform that lets you define a business process, delegate it to an AI agent, and let that agent run it end-to-end with minimal human oversight.

Examples:

A customer success agent that triages support tickets, drafts solutions, escalates edge cases, and closes tickets without human involvement.
A recruitment agent that sources candidates, screens them against JDs, schedules interviews, and sends offer letters.
A contract-review agent that reads NDAs, flags risks, proposes changes, and loops in legal only for novel clauses.

In each case, the agent isn't assisting a human. The human is overseeing the agent.

That's the inversion that makes AOaaS different from Vertical SaaS. And it's why the labor-budget thesis makes sense.

When Do You Choose Agents Over Tools?

This is the million-dollar question every CTO asks. And honestly, it's not always an agent.

You choose agents when:

The process is repetitive and low-variance — Support triage, lead scoring, invoice processing, candidate screening.
Human time is the bottleneck — You have 5 people doing work that could be handled by 1 agent + 1 QA human.
The cost of a mistake is low — Automated triage (human can fix it) beats automated open-heart surgery (human can't).
You have clean process documentation — "We do X, then Y, then Z" is in a runbook. If your process lives in Jira tickets and Slack, agents can't learn it.
The ROI horizon is 6–12 months — If you need labor multiplication next quarter, agents are the move.

You choose tools (Vertical SaaS) when:

The process requires judgment at every step — Executives need tools to make better decisions, not tools to make fewer decisions.
Compliance and auditability are non-negotiable — You need a human signature on every output.
The process is unique to your company — If you're the only company doing it, an agent is probably overkill (unless you're large enough to warrant custom build).

Most mature companies will have both. Tools for judgment-heavy work. Agents for repetitive, high-volume work.

The Bessemer Thesis: Why This Matters

Bessemer Venture Partners published their "Vertical SaaS 4.0" thesis a few years back. The argument was: the next generation of software won't compete on features; it'll compete on business outcomes. Figma didn't win design because it had more features than Adobe; it won because it made designers faster and happier.

The same thing is happening with AOaaS, but one level down. The next generation of Vertical SaaS companies won't compete on features. They'll compete on whether they can run autonomous agents on top of their platform.

Salesforce with agents beats Salesforce without agents, 10 times out of 10. Figma with agents beats Figma without agents. Veeva with agents beats Veeva without agents.

But here's the catch: most Vertical SaaS companies won't build AOaaS layers. They'll integrate a third-party platform (like Astra Space AI) to enable their customers to deploy agents.

So the question isn't "Will Vertical SaaS companies dominate the AOaaS market?" The answer is no. They'll be platforms for AOaaS, not AOaaS themselves.

The actual AOaaS winners will be the ones who build the infrastructure layer that makes it easy for any company — in any vertical — to deploy agents end-to-end, without hiring engineers.

That's Astra's thesis.

How Astra Space AI Fits

Astra Space AI is a platform for the AOaaS era.

We're building the operating system that lets you:

Define an agent in plain English (no code required)
Deploy it to production in minutes
Monitor and improve it via a single dashboard
Integrate it with your existing workflows (Salesforce, Slack, HubSpot, Figma, etc.)
Scale it without hiring engineers

The magic is in the last point. At scale, you can't afford to have your best engineers tweak agents all day. You need a platform where product managers, operations leaders, and even subject-matter experts can build and iterate on agents.

That's what we're building. And that's why we're framing this as an operating system, not a tool.

Operating systems are infrastructure. They're not nice-to-have. They're how your business runs.

FAQ: Everything You've Been Asking

What exactly is AOaaS? Agentic Operating Systems are platforms that enable autonomous agents to handle entire business processes end-to-end with minimal human oversight. Unlike traditional software (which augments humans), AOaaS platforms replace human labor in repetitive, defined processes.

Is AOaaS the same as AI automation? Not quite. AI automation often means simple RPA-style workflows (if X, then Y). AOaaS is different because agents have judgment. An agent can read an email, understand its context, and decide which of 10 different processes to trigger. That's not automation; that's autonomy.

When should I build an agent vs. buy a tool? Build an agent if: The process is high-volume, low-variance, and repetitive (you're paying people to do it today). Buy a tool if: You need to improve decision-making or add features, but humans are still in control.

How do you choose between Astra and a DIY agent? DIY agents are cheaper upfront, but they require constant engineering care. Astra costs more per month but needs zero engineering after the first week. If your labor multiplier is 10x, Astra is cheap. If you only need 20% faster, DIY is better.

Isn't this just ChatGPT with a GUI? No. ChatGPT is a model. Astra is a platform that orchestrates agents, integrates with your systems, monitors performance, and lets non-engineers iterate. It's like saying Excel is just a spreadsheet widget.

What's the difference between Astra and Salesforce agents? Salesforce agents live inside Salesforce and handle Salesforce workflows. Astra agents live outside and can orchestrate across Salesforce, HubSpot, Slack, your data warehouse, etc. Different scope.

How fast can I deploy an agent? Most customers deploy their first agent in a week. Some (with clean process documentation) do it in days. The longest part is usually deciding which process to delegate first, not building the agent.

What about compliance? Can I use agents for regulated work? Depends on the regulation. We have customers using Astra agents in healthcare, finance, and legal. The key is: humans review agent outputs before they go live. That satisfies most compliance frameworks (full audit trail, explainability, etc.).

Is this really a $13T opportunity? Yes. Every company in every vertical has some high-volume, low-variance work. The total labor spend on that work? $13T globally. Our addressable market starts at the top (enterprises with $100M+ ARR and high-volume operations) and expands as AOaaS tooling gets cheaper and easier.

Why is Bessemer's thesis relevant? Bessemer's thesis says the winners in enterprise software are those that deliver outsized business outcomes. AOaaS is the natural extension: not just better outcomes, but labor multiplication. That's Bessemer's vertical-AI thesis. And we're building the platform for it.

What's Next

If you're an investor, founder, or operator reading this: the $13T labor budget is waking up. The Vertical SaaS companies that built the last decade are now integrating agents. The new entrants are building native AOaaS platforms.

The question isn't whether agents will replace labor in your company. The question is whether you'll control that transition, or whether you'll wake up in 3 years and realize your competitor already did.

That's the thesis behind Astra. And that's why we're building it.