The race for synthetic workers: platforms, specialists, and the command layer

A director received two proposals in the same week. A 15-year ERP partner offered a set of AI agents embedded in the system, activatable almost instantly and charged by usage. A specialized startup presented a billing agent that handles the entire process and charges based on results. Both offer the same thing—a colleague who doesn’t take lunch breaks, doesn’t go on vacation, and doesn’t appear on the payroll. The difference lies in who provides the agent.

The army of context

On one side are the owners of registration platforms, where data, rules, and transactions are stored. They argue that no intelligence generates value without context, and that context resides in their databases. SAP has already integrated dozens of ready-made agents into its processes and launched a studio for customers and partners to create their own. Salesforce has transformed CRM into an agent factory, generating $1.2 billion in recurring revenue, with over half of contracts originating from its installed base. It also maintains a marketplace where partners publish ready-made agents. Workday went further: it created a corporate registry of agents, each with a role, permissions, budget, and a shutdown button, and paid over $1 billion for a startup just to provide an interface for this management. The message is clear: the future workforce will consist of humans and agents, and platforms want to be the place where both are managed.

The army of depth

On the other side are specialists—companies that have chosen a specific function or sector and have delved deeply into it. Sierra sells customer service and charges per resolved case. Harvey has become the legal associate for law firms and departments. Hippocratic deploys voice agents to monitor patients. These assets are highly valued by venture capital not because they have more underlying technology, but because they deliver finished work, with engineers embedded within the client. The most advanced companies are already experimenting with result-based pricing and greater operational responsibility. The bet is that depth in one function is worth more than a mediocre presence in several.

Pricing models and the Brazilian experience

While the competition is real, the pricing model is still evolving. The context army charges by usage: Salesforce charges $0.10 per agent action via credits. The depth army charges by results: Intercom charges $0.99 per resolution; in Brazil, Cloud Humans charges between R$ 2.42 and R$ 3.89 per resolved ticket—and scaling to a human is free. Between these two models, a third emerges: managed capacity, where the agent is budgeted as a job position, with quotas, goals, and contracted supervision.

The models are still unstable. Salesforce itself changed its model three times in 18 months: $2 per conversation, then credits per action, then hybrids with a license per user, after customers described the scheme as “a black box that’s hard to budget for.” Consultancies report AI credit contracts signed without clients knowing the usage rates.

Those who switched to result-based pricing report higher retention and expansion—according to market estimates, this is what happened to the model’s pioneer. Customer service and sales converge on results because they are measurable. Platforms converge on usage with governance because, in a generic flow, no one defines “results.” Any current pricing table is perishable, as the cost of operating these models continues to fall rapidly. For buyers, the recommendation is to adopt short pilots, set spending caps, and include renegotiation clauses.

The front line in Brazil

WhatsApp is the arena where the competition unfolds in real time: 82% of Brazilians chat with companies through the app, and 60% have already made purchases through it. Blip, which raised $60 million from SoftBank and Microsoft, projects its attendant will jump from 30 to 300 clients per day with AI support—and XP’s agent on the app has already reduced human activation by 70%. VTEX acquired a startup of agents to integrate post-sales into e-commerce. But the hangover has also arrived: Meetz, a pioneer of synthetic prospecting salespeople, sold its portfolio and changed its business. It’s the same global pattern—Clay, valued at $3.1 billion, won by selling data and sales engineering, not robot salespeople. The lesson from the sales field is clear: qualification, post-sales, and recovery work under supervision. Mass firing has backfired.

How the race ends

The owner of the context has the data, distribution, and buyer trust, but also two weaknesses: conflict with its own revenue—agents that do the work erode the per-user charge that sustained the software for 20 years—and a tendency toward the generic, as platforms need to serve everyone simultaneously. The specialist has excellence, speed, and a client-aligned model, but needs to enter someone else’s context—a barrier that is turning into an opportunity. Open integration protocols now allow external agents to access systems and communicate with each other—even Brazilian ERPs have opened their systems to third-party assistants operating from within. The history of platforms suggests the script: in previous cycles, it was a technical draw. Platforms ended up with volume; specialists, with value.

That’s why I believe this race doesn’t end with a winner—it ends with a division of labor. The generic, low-risk agent becomes a platform resource and is commoditized. High-value, high-risk functions remain with specialists, because depth and responsibility cannot be distributed via a catalog. The most interesting part is the command layer: orchestration within the client. No real company operates in a single system. The actual process spans ERP, CRM, banking, email, and the system of a subsidiary acquired three years ago. An agent tied to its platform sees half the process; a specialist sees deeply but narrowly.

Someone needs to parameterize each worker—skills, rules, limits—integrate them across systems, and govern the fleet with identity, audit trails, and a human manager. This is a space contested by consultancies, integrators, and technology service companies, but it still lacks a natural owner, as no one combines, incontestably, technological independence, knowledge of the client’s processes, and operational responsibility. In Brazil, where the corporate landscape is a mosaic of SAP, TOTVS, proprietary systems, and legacy systems, this is the most valuable position on the board.

Three decisions for the next board meeting

First: apply the context rule to your suppliers. Is your ERP and critical SaaS converting what they know about your operations into agents that work—or are they just charging a license while announcing a roadmap? Don’t buy promises. Ask for the agent in production, with metrics. A supplier who doesn’t turn context into work is asking to be bypassed.

Second: decide function by function, not based on faith in one of the armies. For each operation, there are three options: the platform agent (cheap, integrated, generic), the specialist (deep, result-based pricing), or your own orchestration (for what’s differentiating and crosses systems). One requirement applies to all three: their own identity, explicit limits, an auditable trail—and the guarantee that your business’s data and learning remain yours. Context is your asset; don’t give it away for free on the first demo.

Third: treat any price table as an experiment—because that’s what it is. Use short contracts, spending caps, renegotiation clauses, and constant comparison between usage and results. And don’t wait for the race to end to join it: start with reversible back-office operations, gradually increasing autonomy. Whoever waits for the winner inherits two bad things—the winner’s price and the context already captured by them.

For two decades, choosing software meant choosing functionalities. Choosing suppliers of synthetic workers is a different decision: it’s about choosing who will have access to your business context and who will capture the value it generates. Context, which has always been your greatest asset, has started to work. The question of the decade is for whom.