The companies that win the agentic era will be the ones who own their agent infrastructure, control their own efficiency, and treat agent management as a core operational capability — not a vendor relationship.
When we built Lua, we made a deliberate decision: we never wanted to own your outcomes. We wanted to give you the infrastructure to own them yourself — and manage them with the same intentionality you bring to every other part of your business. We run Lua entirely this way. As we've written about, every person on our team, regardless of function, is expected to build and manage agents as a core part of their job. That's not a quirk of how we work. It's what we believe all successful businesses will look like. And it's a belief that puts us at odds with where the AI pricing consensus has landed. Good. Here's why the consensus is wrong — and why the companies that figure that out first will have an advantage that compounds.
How the pricing consensus formed — and why it's already breaking.
The framework most people use to think about AI agent pricing was built by Madhavan Ramanujam and Josh Bloom at 49 Palms Ventures. It maps two dimensions: how autonomously the AI operates, and how clearly its actions can be tied to a measurable outcome.
The top-right is where the market has converged. High autonomy, clear attribution, pay per result. It made sense. For a while.
Two conditions made it rational:
| — | Buyers didn't fully trust LLMs to deliver consistent quality. Outcomes-based pricing shared the risk — you only pay when it works. |
| — | AI was being benchmarked against a labour budget — but without any of the management discipline that makes labour efficient. The only question being asked was: "Is this cheaper than my current labour costs without impacting quality?" That's a low bar. And it made paying per outcome feel proportionate. |
Both of those conditions are collapsing at the same time. And when they go, the model goes with them.
The next generation of companies don't have a labour budget to optimise against. They're building without it.
The next generation of companies aren't asking "is this cheaper than headcount?" — because they never had the headcount. They're building from scratch with agents as first-class infrastructure. The only question is how efficiently they can run that infrastructure — and whether that efficiency belongs to them or to a vendor.
These companies understand something that the outcomes-based pricing model obscures: agent efficiency is a competitive advantage. It is not a cost to be minimised through the right vendor selection. It is a capability to be built, owned, and compounded — one that separates the businesses that scale from the ones that get outrun by them.
Paying a fixed rate per outcome to a vendor doesn't just fail to build that advantage. It actively prevents it. Every efficiency gain the vendor makes increases their margin. Your cost stays fixed. Your position never improves. You are permanently renting someone else's capability at someone else's price — with no path to owning it yourself.
And there's a deeper problem that rarely gets named: if every company in your market is paying the same rate per outcome to the same vendor, nobody is building an advantage. The efficiency ceiling is identical for all of you. Outcomes-based pricing doesn't just prevent you from improving — it ensures your competitors can't either. Which sounds neutral until you realise it means the only way to break from the pack is to stop paying per outcome altogether.
And right now, somewhere in your market, a small team has already figured this out. They own their agent infrastructure. They've been iterating for months — tightening processes, cutting waste, improving output quality. Every gain belongs to them. It lives in their cost structure and compounds quietly. Their competitors paying per outcome at a fixed rate have no equivalent lever to pull. The gap is already open.
The gap between a company that owns its agent efficiency and one that rents it shows up in margins, in pricing power, in the ability to reinvest. It compounds. And it starts now.
Imagine a business where every function has an agent layer that the team itself owns, manages, and improves.
Where the ratio of humans to agents is a deliberate decision — not an accident of which vendor you chose. Where agent efficiency is a competitive variable that your team moves, quarter by quarter, in your favour. Where the people running sales, marketing, operations, and HR are not just users of AI tools but active managers of an agent workforce they understand deeply and improve continuously.
In that world, the agent layer isn't a black box you pay a vendor to operate. It's a part of the business you own — with all the accountability, institutional knowledge, and compounding returns that ownership implies. Your agents have defined responsibilities. Their performance is measured. When they underperform, you improve them. When a new use case emerges, you build for it. The capability grows with the business because it belongs to the business.
This is what human-agent collaboration looks like when it reaches its potential. Not a chatbot handling overflow tickets. Not a vendor-managed automation running in the background. A real operational layer — built intentionally, managed rigorously, and owned entirely by the team it serves.
Agent costs will scale like labour costs. The businesses that manage them like labour will win.
That future is available to any business willing to build toward it. But it requires a decision — and the window to make it is narrower than most people realise.
Agent headcount is already a deliberate strategic decision for the most advanced operators. As that becomes true for the mainstream — and it will, fast — the businesses that have been building that capability will have a structural advantage that is very hard to close. The ones that have been renting outcomes will have a cost structure, a dependency, and a capability gap to deal with all at once.
The compounding gap isn't just a cost story. It's a capability story. The companies building their own agent infrastructure aren't just getting cheaper — they're getting smarter. They understand their agents in a way that a company paying per outcome never will. That knowledge doesn't exist on any vendor's platform. It can't be purchased. It has to be earned through the work of building, deploying, observing, and improving.
Some companies will always want to buy solutions — and outcomes-based products will serve that market. But the companies that define the next era won't be among them. They'll be the ones who decided early that managing agents was a core operational skill, not a procurement decision. The ones who hired for it, built for it, and treated it with the seriousness it deserves.
"When I talk to the companies getting the most from Lua, they describe us the way they'd describe an HR function for their agent workforce — not a tool they use, but the infrastructure that helps them build, manage, and develop the agents that run their business."
The future of work has an agent team in it. The question is who manages it.
We're moving toward a world where every function of a business has an agent layer, and the people running those functions are expected to build, manage, and improve that layer as a core part of their job. Not as a technical skill. As an operational one. The question isn't whether agents become central to how businesses work — they already are for the most advanced operators. The question is whether your business is the one shaping that layer, or paying someone else to run it.
If the market settles into a model where most businesses rent outcomes from vendors rather than building the capability to generate them, we'll have failed to capture what this technology actually makes possible. We'll have turned the most significant shift in how work gets done into another SaaS line item.
The beauty of human-agent collaboration — when it reaches its actual potential — is that people and agents working together with real intentionality can build organisations that operate at a scale and quality that neither could achieve alone. That only happens when the humans are genuinely managing the agents. Understanding them. Improving them. Owning the outcomes they produce.
That's the world Lua is building toward. A platform that gives any business the infrastructure to own their agent workforce — to build it, deploy it, and make it more efficient every quarter, so that every gain belongs entirely to them. Not to us. Not to any vendor. To the people doing the work.
See how Lua works →
