Written by Alexander Ekdahl

As AI keeps getting smarter, the old way of buying software by sticking to one vendor’s platform just doesn’t make sense anymore. When digital colleagues can handle tasks across different systems without a human lifting a finger, it’s the result that matters, not the logo on the app. Instead of thinking about which brand the AI comes from, it’s time to focus on how well it gets the job done. In this new world, the smartest move isn’t buying software that happens to have an AI agent, it’s picking the agent that actually delivers the best outcomes, no matter where it comes from.

In a series of articles, Algorithma has dived into organizational, technological and processual impact in businesses. In this article we deep-dive into how to procure - or rather recruit - digital colleagues to your teams.

*“Buying an agent because it’s native to a platform is like hiring someone just because they live near your office. In the end, it’s the outcome that matters.”*

- Alexander Ekdahl, AI Leader

A silent fork in the road

Agentic AI is booming, and every major software vendor is racing to bolt an AI agent onto its existing product line. That instinct is predictable: new technology arrives, incumbents wrap it in familiar packaging, and customers assume the safest choice is to stay inside the ecosystem they already know.

But this time the stakes are very different. The more capable these agents become, the less their value is tied to a particular interface, data model, or vendor brand. An agent can draft the contract, file the purchase order, or onboard the employee without touching the application.

As these agents learn your preferences and collaborate across departments, they begin to feel less like software features and more like digital colleagues, workers who inhabit the flow of business rather than the screen of an app.

The standard playbook: Wrap the new around the old

Major software vendors have been quick to bolt AI agents onto their existing product lines, adding agentic capabilities across office tools, CRM systems, and collaboration platforms.

On the surface, these updates promise smarter documents, faster workflows, and better integration. But under the hood, much of the value still comes from large language models calling the same APIs that were already available just without forcing users to navigate outdated interfaces.

The fallacy of tool‑locked agents

Vendors insist their agents integrate best with their own platforms which, technically, is true. Yet that logic only matters if the platform stays at the centre of gravity. In an agent‑first world, the centre of gravity shifts to:

When your digital colleague can jump from one application to another as easily as you switch browser tabs, buying an agent solely because it’s “native” to a platform begins to look like hiring talent because they happen to live near your headquarters.

The task, not the tool, becomes the product

Once the agent is the primary interface, the underlying software is just another service layer, an API endpoint your digital colleague pings when it needs storage, business rules, or a transaction signature. That inversion has two profound consequences:

Vendor lock‑in takes a new form
Lock‑in used to mean migrating data out of a database. Now it means migrating prompts, task graphs, and learned behaviours that span multiple systems.
Software obsolescence accelerates
If an agent can switch from SAP to NetSuite with a different connector, the traditional moat of “we’re already installed” erodes quickly.

What happens next: A freer, task‑first future

As we shift into the agentic era, the rules of competition change. Success won’t come from simply embedding an agent into existing software, it will come from creating digital colleagues that excel at specific outcomes, no matter what platform or tool they tap into. Here are the types of agents that will define the next phase:

Broadly, three types of agents are emerging as the most effective models, each offering distinct advantages depending on the organisation’s goals and technical landscape.

Domain-specialist agents

These AI agents have deep expertise in specific areas such as legal, finance, procurement, or supply chain operations. Rather than relying on shallow “native integrations” that only touch the surface of workflows, these agents master the real process steps, edge cases, and complexities. They excel at handling nuanced, end-to-end tasks with a level of skill that generic assistants can’t match, becoming trusted digital colleagues in areas where precision and domain knowledge are critical.

Interoperable agents

These AI agents are designed to move freely across systems. They treat SaaS platforms, databases, and cloud apps as interchangeable utilities, plugging into whatever system is required without remaining loyal to a specific vendor.

These agents avoid becoming trapped inside a single platform’s ecosystem. Instead, they stitch together information, actions, and decisions across multiple platforms, creating a seamless experience where the task, not the tool, drives the workflow.

Autonomous, end-to-end agents

These go a step further by handling entire workflows independently. From gathering information to making decisions to executing actions, these agents work without constantly requiring human oversight. They dramatically reduce cognitive load for humans and act not as glorified assistants but as self-directed teammates. This allows them to take tasks from start to finish with minimal interruption, freeing human colleagues to focus on higher-value, strategic work.

Procurement philosophy for the agentic era: Don’t buy software that merely includes an agent. Hire the digital colleague that best delivers the outcome. The rest is plumbing.

Portability as a Strategic Consideration

As enterprises move toward outcome-first procurement, choosing the right digital colleague requires a new evaluation lens: portability. Traditional software selection focused on features, vendor reputation, and integrations. But in an agent-first world, what matters is how easily an agent can adapt, connect, and scale across the organisation’s diverse systems and workflows.

To support this shift, a simple indicator of agent portability can be considered. While the full assessment methodology can vary by industry and use case, agent portability typically considers factors like:

System flexibility: the agent’s ability to operate across multiple platforms and databases.
Connector and API openness: support for open standards and ease of integrating with new tools.
Memory and learning portability: how easily learned behaviours, prompts, and task graphs can transfer between environments.
Degree of vendor dependency: the extent to which the agent’s core functionality is tied to a single vendor’s ecosystem.

An agent with a higher agent portability is not only more adaptable today but also more resilient to future changes in technology, vendor strategies, and business needs. While detailed scoring depends on the specific requirements of each organisation, the principle is universal: the more portable the agent, the lower the long-term switching costs and the higher the strategic flexibility.

A call to rethink enterprise strategy

When talent went remote, companies stopped hiring by ZIP code. As intelligence goes agentic, enterprises will stop buying by platform. The sooner we embrace that shift, the sooner we unlock the full potential of AI without recreating yesterday’s silos in tomorrow’s workflows.

Winning in the agentic era means leaving behind legacy habits for good. Start by designing workflows around outcomes, not around apps or vendors. Prioritize agents that can adapt, connect, and deliver value across any system. Build tech stacks that are modular, flexible, and ready for change. Stop asking, “Which platform do we buy?” and start asking, “Which agent gets the job done best?” The companies that move fastest won’t be the ones with the biggest software contracts, they’ll be the ones with the smartest, most autonomous digital colleagues leading the way.

Buying AI by vendor? That’s cubicle‑era thinking