Managing Two Curves at Once: How Software Companies Lead the Shift to an Agent Business

In short: In the years ahead, software companies have to manage two curves at once: the existing business, which has to stay stable because it funds the transformation, and the new agent business, which needs to grow. That works with a separate view of both businesses, staged funding and a narrative that ties the two together. Vendors in the German-speaking region have an often overlooked advantage on top of that: trust can become a product feature.

Why is the transformation a leadership task?

The Bitkom study “Softwarewelt 2036” explicitly describes AI as a job for top management, not a topic that can be delegated to IT. The executives surveyed see the biggest brake in people, leadership and culture, far more than in technology. Technology alone delivers no value if the organizational model, incentives and skill profiles don’t move with it.

That matches what I see in practice. A new pricing model founders on sales compensation, an in-house agent on cost control, a new product on an organization built for the old one. Only the executive team as a whole can see these dependencies. That includes using the tools themselves.

If you only know AI from slide decks, you’ll consistently underestimate how much is changing.

How do you run the old and the new business at the same time?

The software industry has handled this task once before, in the move from installed software to subscriptions. Some vendors reinvented themselves back then. Others protected their license revenue and fell behind. The difference today is speed. With AI, new features are built faster and more cheaply, the barriers for challengers are lower and the window for established vendors is shorter.

Managing two curves takes three things.

A separate view.Measure the new business against the metrics of the old one and you’ll shut it down too early, because it costs money at first. Measure the old business against the expectations of the new one and you’ll neglect the foundation.

Staged funding.Some now, some once clear milestones are reached. That guards against betting everything on one card and against starving the new business.

A narrative that ties it together.People in the core business need to understand why their work still matters. People in the new business need to understand why they depend on the core business. Without the data from the existing business, there is no good agent.

Why is trust becoming a competitive advantage?

European regulation is often described as a brake. I think that view is incomplete. Regulation raises the bar for everyone. A fast-growing challenger has to meet it after the fact, often under time pressure and with gaps. An established vendor that has worked with regulated customers for years has processes, documentation and experience it can build on.

From the industry’s perspective, the Bitkom study reaches a similar conclusion. Compliance, data sovereignty and regional market knowledge can become differentiators for German and European vendors. It calls the principle compliance by design, meaning compliance as part of the architecture from the start. For vendors selling agents, that translates into four verifiable product features:

  • Where processing happens: data and processing by the agent can stay in Europe, even if that costs more.
  • Traceability: every action, data query and approval by an agent is logged and justified in a way an auditor can read.
  • Choice of model: language models are interchangeable, so customers with special requirements can use a European or self-hosted model.
  • Operational reliability: commitments on availability, contingency plans and a fallback to manual processes if an agent fails.

How much autonomy should an AI agent get?

Autonomy should always be granted explicitly. You need to define where an agent may act on its own, who approves that and how often these rights are reviewed. Software vendors should give their customers the tools to make that call themselves. For many operational areas, I find three levels workable:

  1. Level 1: The agent proposes and explains, a human approves.
  2. Level 2: The agent handles routine cases itself and flags exceptions for approval.
  3. Level 3: The agent acts on its own within defined limits, and humans review after the fact.

These settings have to sit with the customer, be documented and be changeable at any time. A vendor that gives its customers this control gives something up and gains something more valuable: their willingness to use agents in the first place.

Who owns what an AI system learns at the customer?

Every correction, every exception and every workflow an agent processes generates knowledge about the customer’s organization. Gartner explicitly advises established vendors to capture and retain this customer-specific knowledge, not just the data (Gartner, July 1, 2026). For customers, that’s also a concern: will their hard-won know-how flow into a model their competitors use too?

A clear separation holds up. What an agent learns about a specific customer, its rules and exceptions, stays assigned to that customer, improves only its workflows and can be taken along or deleted if the customer switches vendors. What the vendor learns from anonymized patterns across many customers benefits everyone, with explicit consent and the option to opt out. I consider this one of the most important contract questions of the coming years. The legal details belong in the hands of specialized lawyers.

What does a 100-day plan for the transformation look like?

  1. Days 1 to 30, understand the situation: test the load-bearing assumptions of your business model, first by gut feel, then with numbers. Talk to your three largest customers about their plans for agents, to your investor about their expectations, to experienced employees about their concerns.
  2. Days 31 to 60, set the direction: infrastructure, your own agent layer or orchestration, and in what combination. From that follow the outlines of the pricing model, contract clauses for agent access and learned knowledge, and a first small team with a clear mandate.
  3. Days 61 to 100, pilot and communication: a pilot with two or three customers whose data is good enough. In parallel, communication with the investor, the workforce and existing customers. At the end, a second review of whether the picture has changed.

A pilot is the start of execution, not a substitute for it. The plan for the time after the hundred days therefore has to take shape during the third phase.

Which metrics show whether the transformation is making progress?

The classic SaaS metrics remain important, but they aren’t enough. I suggest five additional metrics that most companies can collect within a few weeks:

AreaAdditional metricWhat it shows
ProductShare of revenue from offerings based on hard-to-copy data and industry expertisewhether the company gets paid for what is hard to copy
Revenue modelShare of recurring revenue not tied to usershow far the model has moved away from the seat
Customer relationshipAcceptance rate of suggestions from your own agentswhether customers trust your own agents
MarginGross margin of the agentic layer, separate from the rest of the businesswhether the new business is growing profitably
OrganizationShare of employees who regularly use AI tools in their core workwhether the organization can carry the transformation

The trend across quarters matters more than the exact value. With these numbers, an executive team can show investors, the advisory board and the workforce where the company stands.

Which conversations can’t an executive team delegate?

With investors: valuations are shifting from pure revenue growth toward operating profit. An executive team has to show that it understands the new costs, keeps the core business stable and has a path to profitability for the new business. It pays to find out early whether your investor thinks through the cycle or looks at the next quarter.

With the workforce: employees want to know whether the company will make it through the change and what it means for them personally. The first question can be answered at an all-hands meeting, the second only in one-on-one conversations. In Germany, the works council needs to be involved early.

With existing customers: they fund the company, provide the data the agents build on and are the first to decide on a new pricing model. Openness about what isn’t working yet earns more credit than promises. And the most important customers are often further along than their vendor assumes.

Frequently asked questions

Who should own the AI transformation in a software company?

The executive team. The transformation affects product, pricing, sales, costs and organization at the same time. Those dependencies can’t be managed from within a single department.

Is European regulation a disadvantage when deploying AI agents?

Not necessarily. Vendors that build in traceability, data sovereignty and human oversight from the start open the door to customers in regulated industries who wouldn’t buy without that proof.

How long does the shift to an agent business take?

The first decisions can be made in about a hundred days. For most mid-sized software companies, the transformation itself takes several years. What matters is starting early, because the window for established vendors is shorter than it was in the move to subscriptions.

Further reading: Is SaaS dead? Why that’s the wrong question · From tool budget to headcount budget: pricing for AI agents · Who owns the customer when the agent does the work?

Sources: Bitkom (2026): Softwarewelt 2036 · Gartner press release, July 1, 2026

Similar Posts