Agents-as-a-Service: the new platform shift

NVIDIA CEO Jensen Huang took the stage at GTC 2026 and told the world the AI era had officially arrived. The line that landed hardest: every engineer at NVIDIA should receive an annual token budget worth roughly half their salary. If a $500,000 engineer wasn't consuming at least $250,000 in tokens, Huang later said he would be "deeply alarmed."

He described tokens as "the new commodity" and rebranded data centers as token factories, laying out a ten-year vision in which 75,000 NVIDIA employees would manage 7.5 million AI agents. One hundred agents per person. And one prediction in particular reframed the future of enterprise software:

"Every SaaS company will become an agents-as-a-service company."

Agents-as-a-service: the new SaaS?

We think yes. But before we jump too far into our findings on the topic, let's set some context.

For two decades, the SaaS model was built around structured databases and fixed workflows. These tools required humans to log in, click buttons, and manually perform the labor.

Agents-as-a-Service reimagines that relationship. Instead of a singular tool housing isolated data, an agent connects multiple tools and reasons across data to finish a task. A simple way to think about the difference:

• Software-as-a-service = structured data and manual workflows
• Agents-as-a-service = connected data and autonomous outcomes

In this new model, employees go beyond using software to collaborating with it. Agents function as digital coworkers, equipped with specific job responsibilities and access to systems in order to deliver results autonomously.

Coincidentally, we spoke with 10 different AI enablement managers the week of the GTC keynote: AI operations leads, change management consultants, and strategy advisors at companies actively rolling AI out across their workforces. A consistent picture emerged from those conversations.

AI Adoption is leaking Platform Engineering vibes

Much like the last 10 years in R&D organizations, the lines between internal AI development, operations, and adoption are extremely blurred. And just like DevOps and platform engineering leaders, every person has essentially the same job, and almost none share a title. One led AI adoption and engagement. Another is a Manager of AI Operations. Another is an AI Enablement Manager. One runs AI Strategy. One person built an entire internal AI ecosystem from scratch at a company you wouldn't expect. All served as internal leaders serving their company's adoption and success with AI. These roles are being created independently across industries, sitting between the engineering team and the rest of the company and serving the company's employees as their own internal customers. Companies are putting real headcount and investment towards making AI land for every employee, starting with R&D.

The bottleneck is context and trust

An AI adoption lead at a cybersecurity company, two years into rolling out ChatGPT, Copilot, and Claude inside his organization, summarized the issue in one line: most employees don't know how to give AI the context it needs to be useful. Without context, the model summarizes. With context, it thinks. An AI operations leader at a financial compliance company described the same gap from a different angle: his team had a well-modeled data lake, a semantic layer, and a BI tool, and the AI still couldn't answer the questions leadership actually cared about until it understood segmentation rules, customer lifecycle stages, and how the company defined churn. The pattern shows up everywhere: 88% of companies report regular AI use, but only about 6% are capturing meaningful ROI. The problem isn't the AI. It's that most AI today operates without the context it needs to deliver business value. (We wrote more on this in Why 95% of companies aren't getting value from AI.)

Company-wide memory is missing

A change management consultant working with large enterprises described what she sees across her client base: companies moved fast into AI, bought licenses, and are now discovering that adoption is uneven. A handful of power users are getting enormous value, but everyone else is still figuring out where AI fits into their day. She shared that the way AI is being packaged and delivered to most employees hasn't caught up to what the technology can actually do.

An AI strategy lead at an 800-person developer tools company described the day-to-day experience as "read, then manually act" across one tool for search, another for tasks, another for documents, with no intelligent layer connecting them. About 20% of the company uses Claude Cowork today, and the ambition is much higher. The piece that's missing is connecting AI to the full picture of how the business actually operates - the "company brain."

Two worlds, same playbook

For developers, the transformation has already happened. 92% of US developers use AI coding tools daily, and 41% of code globally is AI-generated - and climbing every day.

We've been learning for the last two decades that developers quietly run the technology industry - and AI tools are smartly built for how developers actually work. This puts the power directly into developers' hands, heavily contributing to this transformation moving faster than any other technological shift we've seen.

The rest of the workforce is next, and the right playbook will follow the same principles: connect AI to the tools people already use, give it the context it needs, and meet them where they spend their time.

The models can already reason at an extraordinary level. The next step is connecting them to the context they need to reason about a specific business, and delivering them in a way every employee can use.

Agents-as-a-service architecture

Three layers work together to form agents-as-a-service architecture.

Models. Today's models can handle multi-step tasks, call external tools, execute code, and reason across a large amount of company data. (We cover more on the latest model capabilities in AI for Startup Leaders.)

Connected data. Unlike SaaS, which locks your data inside a single application's database, an agents-as-a-service architecture connects your full suite of tools. This includes your CRM, project management, code repositories, communication tools, payment systems, data warehouses, and more.

UI surfaces. Agents meet you where you already work. That might be Slack, where you @mention an agent the way you would a colleague. Or it might be a web interface. The agents-as-a-service model does not require you to log into another dashboard. The agent shows up in the channels and tools your team already uses.

This architecture enables a network of sub-agents to work together, coordinating or "swarming" to deliver high quality work, essentially handling the tough parts of agent orchestration for you.

Agents-as-a-service examples

Think of agents as having job responsibilities, and working autonomously to help a team. Here is what that looks like in practice:

• A sales rep needs to prepare for a prospect call. So an agent pulls context from the CRM, recent emails, and LinkedIn, then assembles a briefing. After the meeting, the agent drafts a follow-up, updates the deal record, and flags risks to the pipeline.
• A finance leader notices bookings are soft this week. So an agent checks Stripe, HubSpot, and related systems, traces the likely cause, and delivers an explanation within a #leadership Slack channel with supporting data. Along the way, it flags churn risk and surfaces anomalies before they show up in a dashboard.
• A marketing team launches a campaign and needs to know what is working. So an agent writes and schedules content, tracks performance across platforms, and reports on CAC trends by channel. It connects attribution data to revenue so the team can see which efforts actually moved the needle.
• An engineering manager needs a clear view of delivery status. So an agent cross-references Linear issues with GitHub pull requests, identifies stalled work, and posts weekly updates to Slack. It keeps systems synchronized and flags when something is falling through the cracks.

Each of these agents operates across multiple tools, like employees you collaborate with to complete a job. This is why Jensen Huang is excited about agents-as-a-service. Under the old model, adding another SaaS in your business led to sprawl. However, adding an additional agent can help teams produce faster, higher quality work.

Adapt is an agents-as-a-service company

Adapt is a universal AI agent that connects to your full work stack, including HubSpot, Slack, Linear, GitHub, Stripe, Google Workspace, BigQuery, and anything with an API. It learns your business context over time, building organizational memory that benefits the entire team.

Our founding team all have backgrounds in building developer products and understand the right playbook well. For example, Adapt operates where your team already works, instead of forcing users to use a generic dashboard. Users can tag @Adapt in any private or shared Slack channel or thread or in the web app. The agent shows up, understands the context of the conversation, and takes action. You can set scheduled tasks or proactive policies for the agent, too, so you can focus on building loops for the agent instead of one-off prompting.

The results are measurable. DoNotPay's CEO reduced a 45-minute customer support task to under one minute. Wander's CTO deflected data requests away from the engineering team, freeing them for strategic work. And RevSend's founder eliminated ad-hoc reporting interruptions entirely - they run 100% of their RevOps workflows on Adapt.

These startups are living in the future that NVIDIA's CEO predicted. They are growing by replacing SaaS tools that employees operate, with agents they can delegate to.

Try out Adapt with your team and get $100 free credits

FAQ

What is Agents-as-a-Service?

Agents-as-a-Service is a new software model where AI agents are deployed to fulfill job responsibilities across multiple tools and data sources. Jensen Huang helped popularize the term at GTC 2026, predicting every SaaS company will make this transition.

How is Agents-as-a-Service different from SaaS?

SaaS delivers software organized around structured data and workflows. However an Agents-as-a-Service is organized around connected data and outcomes. Work is delegated to an agent and it reasons through data and uses tools until it completes a task.

What does agents-as-a-service architecture look like?

An agent service usually includes three layers: models, data sources, and UI surfaces. This architecture enables swarms of agents that can coordinate across systems to deliver outcomes no single tool could produce alone.

What are examples of agents-as-a-service?

Agents function as role-based coworkers. They handle sales prep and follow-ups, monitor and explain business performance, run and measure marketing campaigns, and track engineering execution in collaboration with human teammates.

What companies offer Agents-as-a-Service?

Adapt is an agents-as-a-service company, providing a single AI agent that connects to your entire tech stack and works where your team already works.