We are not building another enterprise AI agent company. We are rethinking how businesses operate, learn and get things done.
For decades we have built increasingly sophisticated software to help businesses run. ERP systems, CRMs, accounting platforms, project management tools and countless specialized applications have become essential to everyday operations. They have helped us organize information, standardize processes and manage growing complexity. Yet despite all this progress, people still spend an extraordinary amount of time operating software rather than focusing on what their businesses are actually trying to accomplish. We open applications, fill out forms, update records, check dashboards, chase emails and piece together information scattered across different systems. We have become remarkably good at building software that records what happens inside a business. I am not convinced we have become nearly as good at building software that helps the business get things done.
I think AI gives us an opportunity to fundamentally change that relationship. For the first time we have technology capable of interpreting context, reasoning across complex situations and participating in the work itself rather than simply recording it. We can begin imagining systems that understand what a business is trying to achieve, recognize what is standing in the way and help people do something about it. But realizing that potential requires us to question some of the assumptions that shaped enterprise software in the first place.
What concerns me is that we may be repeating the same mistakes with AI that created today’s fragmented enterprise software landscape. Consider an AI customer service agent that can answer questions, respond to complaints and resolve support tickets but has no real understanding of the business processes behind those interactions. It might be excellent at handling conversations but have no idea whether the company can actually fulfill what it is promising the customer. How is that fundamentally different from the disconnected systems we already have? We have added intelligence to individual applications without necessarily making the enterprise itself any more intelligent. An extraordinary amount of capital, engineering effort and resources is being spent attaching AI capabilities to existing software and building specialized agents around the same functional boundaries. Some of these tools will undoubtedly be useful, but we risk recreating the very fragmentation we should be trying to eliminate. We could end up with hundreds of intelligent agents that are individually capable but collectively no better at running a business.
A Business Is More Than Its Software#
Take a manufacturer for example. At the end of the day the goal is relatively simple. Build a great product and deliver it to the customer. But anyone who has actually run a manufacturing business (ask me how I know) understands just how much has to happen behind the scenes to make that possible. Materials have to arrive, suppliers have to fulfill their commitments, production needs to stay on schedule, quality issues need to be resolved and customers need accurate information about their orders. Finance needs to support everything happening along the way. Hundreds of interconnected processes, decisions, dependencies and people have to come together to deliver that one finished product.
And that’s assuming everything goes according to plan, which it rarely does. A supplier misses a shipment, a component fails inspection, a purchase order gets delayed or a customer changes a requirement halfway through production. Suddenly a seemingly straightforward commitment becomes a complicated exercise in coordination, judgment and problem-solving. Someone has to understand what happened, figure out what it affects, identify possible solutions and bring the right people together to keep things moving. Often the information needed to make those decisions is spread across emails, spreadsheets, ERP records and conversations that happened days or weeks earlier.
What I find fascinating is that great businesses have learned how to navigate this complexity. They have spent years and sometimes decades developing the knowledge, judgment and relationships required to deliver consistently despite everything that can go wrong. That accumulated experience is one of the most valuable things a business possesses. Yet much of it remains in people’s heads, conversations and disconnected systems. Our software captures individual transactions and activities but rarely preserves the reasoning behind decisions or the lessons learned from experience.
That is the gap we want to close.
What If the Enterprise Could Learn?#
At Delegus we are building toward what we call the learning enterprise, a business that is human-led, agent-powered and governed by bounded authority. The idea is that an enterprise should develop an understanding of how it operates, how work moves between departments and what needs to happen to fulfill its commitments. Rather than waiting for someone to ask the right question, it should recognize emerging problems, connect the relevant information and help people decide what to do next.
Imagine a production manager whose team is working toward several customer deliveries. A critical supplier shipment is delayed and the information appears in an email. In a learning enterprise the system could recognize that the delay threatens three deliveries, identify possible alternatives and explain the situation with the relevant context already assembled. The manager could then ask it to confirm the revised delivery date, investigate alternatives and prepare a recovery plan. AI agents would gather information, coordinate with the appropriate systems and bring back decisions requiring human judgment.
The important change is that the manager no longer needs to discover every problem or personally coordinate every step of its resolution. The enterprise can help recognize what matters and carry out the work that follows. People remain responsible for the decisions while agents take on more of the execution.
A New Operational Fabric#
Making this possible requires more than adding intelligence to individual applications. Businesses do not operate in the neat categories our software was designed around. A customer order represents a commitment that may involve purchasing, production, finance, logistics and customer relationships. Each function has its own systems and responsibilities but the customer ultimately cares about one thing. Did the company deliver what it promised?
We believe the next generation of enterprise software needs a new operational fabric that connects people, knowledge, systems, decisions and actions around those outcomes. Rather than replacing the applications companies already depend on, this fabric should connect to them and develop an understanding of the relationships between the work happening across the organization. It should understand dependencies, recognize when something changes and help coordinate the actions required to keep the business moving.
The important distinction is that the outcome becomes the organizing principle rather than the application or the process. Instead of treating purchasing, production and logistics as separate worlds, we can begin to understand how they contribute to delivering a finished product. That shared context allows intelligence to operate across the business rather than remain trapped within individual functions.
And it changes how people should interact with enterprise software.
From Operating Software to Making Decisions#
For decades we have trained people to operate software. We teach employees which applications to open, where to click, what information to enter and which reports to generate. In many organizations a significant part of someone’s day is spent keeping software synchronized with what is happening in the real world. I think we should reverse that relationship and start building software that understands enough about the business to work naturally with people.
Imagine a contextual interface where someone can talk, type or interact visually with their business through the perspective of their own responsibilities. A purchasing manager understands supplier commitments and material availability. A production manager sees capacity constraints and delivery risks. Someone in finance understands cash requirements and financial exposure. A business owner can see how these activities affect the company’s overall performance. Everyone works from the same underlying understanding of the business but interacts with the information and decisions that matter to their role.
A production manager might simply ask whether the company is on track to deliver everything promised this month. Instead of receiving another report to interpret, the system could explain what needs attention and offer to handle the next steps. The manager could delegate follow-ups, request alternatives or ask for a recommendation without manually navigating the underlying applications or deciding which AI agent should perform each task.
This is not about removing people from the process. It is about changing where their time and attention go. People should spend less time searching for information, coordinating routine activities and updating systems. They should spend more time understanding situations, exercising judgment and making decisions that move the business forward.
We want to move people from operating software to directing outcomes.
Intelligence Is Not the Same as Authority#
As agents become capable of performing meaningful work, an important question emerges. Who decides what they are allowed to do? An agent might be capable of negotiating with a supplier, changing a purchase order or initiating a payment. But capability alone does not mean it should have permission to perform those actions.
At Delegus we are building around the concept of bounded authority. People should be able to delegate work while clearly defining what an agent can do, what it cannot do and when it needs to return for approval. A purchasing agent might be allowed to contact suppliers, request quotations and compare alternatives. It might be authorized to place orders within a defined financial limit but require explicit approval for larger commitments. These boundaries cannot simply exist as instructions in a prompt. They need to be enforceable, revocable and independently verifiable.
This becomes especially important when agents interact across organizational boundaries. Imagine an AI agent representing a supplier requests a change to the bank account used for payments. How does the manufacturer receiving that request know the agent genuinely has authority to make it? Knowing which agent sent the request is not enough. The receiving organization needs to determine whether that specific action was authorized by the appropriate party and whether that authority is still valid.
That is one of the problems we are addressing with Delegus Check. We want organizations to be able to verify delegated authority before consequential actions take place and retain verifiable evidence of what was authorized and what happened afterward. The objective is not to give AI unrestricted freedom. It is to make delegation trustworthy enough that businesses can confidently allow agents to take on meaningful responsibility.
A Business Should Get Better Because It Has Operated#
There is another dimension to this vision that I find particularly interesting. Businesses learn through experience every day. A purchasing manager discovers a more reliable supplier. A production team finds a better way to organize its work. Someone in customer service figures out how to resolve a recurring issue. A business owner makes a difficult decision, learns from the outcome and changes how the company approaches similar situations in the future.
Unfortunately much of that learning never becomes part of the organization’s lasting knowledge. People leave, conversations disappear into email archives and processes change without anyone recording why. The same mistakes are sometimes repeated because the lessons were never shared beyond the people directly involved. Our systems are reasonably good at preserving transactions but much less capable of preserving the experience behind them.
A learning enterprise should help change that. It should recognize recurring patterns, understand which interventions produced better outcomes and suggest improvements to how work gets done. When someone corrects a recommendation, explains a decision or chooses a different approach, that feedback should improve the system’s understanding of the business. Over time the organization should become better at anticipating problems, evaluating alternatives and delivering its commitments.
Learning should not mean silently expanding an agent’s authority. A system can become more capable through experience while the boundaries governing its actions remain under human control. That distinction allows the enterprise to improve continuously without losing the accountability that makes businesses trustworthy.
The Opportunity Is 10x Better Outcomes#
I keep coming back to a simple question. What would it actually mean for a business to become ten times better at delivering what it promises? Not ten times more dashboards, ten times more automation or ten times more AI agents performing isolated tasks. I mean genuinely better business outcomes.
For a manufacturer that could mean dramatically fewer missed deliveries, faster resolution of supply chain problems, better use of working capital and far less time spent coordinating information between departments. For a services company it could mean completing customer commitments with less administrative overhead and fewer handoffs. For a growing business it could mean increasing operational capacity without adding the same amount of complexity every time the company gets bigger.
These are ambitious goals and we will have to demonstrate them through real deployments and measurable results. But I believe this is the scale of improvement we should be pursuing. Making someone 10% faster at completing an administrative task is useful. Helping an entire business become dramatically better at fulfilling its commitments is something else entirely. The opportunity is not simply to automate more tasks but to increase what the organization is capable of accomplishing.
And perhaps the most important part is what that means for people. If intelligent systems can take on more of the coordination, information gathering and routine execution, people can devote more of their time to the things that benefit from human experience, creativity, relationships and judgment.
Human-Led, Agent-Powered and Continuously Learning#
We are still early in understanding what businesses will look like when AI can participate meaningfully in their daily operations. I don’t believe the answer is to replace every application with a chatbot or deploy hundreds of agents and hope they figure everything out. Businesses are complicated for good reasons. They are built around people, commitments, relationships and decisions that often carry real consequences. We should respect that complexity rather than pretend it can simply disappear.
I imagine a world where a business owner can connect the applications their company already uses and begin understanding what is happening across the organization without needing a developer or a team of consultants. Where someone in purchasing, production, finance or customer service can interact naturally with their part of the business through conversation, voice and a contextual interface. Where important work is recognized early, the right people are involved in decisions and agents can carry out authorized actions across the systems and organizations involved.
That is what we are working toward at Delegus. We are not trying to build software that simply helps people manage more work. We want to help businesses accomplish more of what they exist to do while preserving the knowledge, judgment and human responsibility that make them successful.
Humans lead. The enterprise learns. Agents do the work. Authority stays with people.
Footnote — Claude contributed to the drafting guidance, structure and formatting of this article. The ideas and perspectives are my own.