Over more than two decades of building and scaling revenue organizations, I’ve seen the same pattern too many times. Each GTM tool promises to give salespeople more time with customers. Over time, these tools collect and create fragmented islands of data. Sales reps and RevOps spend more and more of their time reconciling conflicting information and tracking down the information they need.
Tools and technology was supposed to make this easier, but a Coveo study last year found that nearly half of employees in large organizations said the information they need is stored across multiple applications and they spend an average of three hours per day searching for it.
At one company where I led the sales organization, reps used roughly a dozen applications and browser tabs, often entering the same information repeatedly. Every tool addressed a legitimate need, but together they created an expensive mess that ultimately reduced time spent with customers.
Advertisement
AI now makes it possible to repeat this mistake faster and at scale.
AI agents are all the rage. Proponents tout their ability to let front-line workers address their own needs without waiting months for IT or paying a vendor for half a solution. B2B ratings and reviews platform Clutch recently reported that nearly two-thirds of full-time workers it surveyed have built their own AI agent.
Building agents can be rewarding. A salesperson who spends Friday afternoon creating an agent to prepare for Monday’s customer feels productive. The agent works, and the meeting brief looks useful. But if ten colleagues also spent their Friday afternoons solving essentially the same problem, is the organization really any better off?
Giving people the ability to build software can be valuable, but recreating routine sales tools at scale can consume the very time AI was supposed to give back. There are also security risks. Clutch found that 91% of employee-built agents can access company data. Gartner predicts that more than 40% of enterprises will experience security or compliance incidents linked to unauthorized shadow AI by 2030.
This looming problem gets too little attention because, on the surface, the impact appears benign. Agents can be spun up in a couple of hours. Each individual project looks inexpensive, and the scope is usually limited to a few processes or people.
But what looks like low-cost experimentation at the individual level can become a costly and inefficient operating model at scale. Now imagine replacing those overlapping subscriptions with overlapping agents. One seller builds an account research assistant. Another creates a meeting preparation tool. Revenue operations codes a prospecting workflow that duplicates parts of both. Which is the right tool for everyone to use?
In most cases, it’s all three. The software-as-a-service industry has a term for this: “SaaS sprawl.” AI will make the problem much bigger and harder to manage.
Costs go well beyond writing code. Someone must test the output, maintain integrations and respond when a model or data source changes. Access permissions must be maintained. Creators take on a maintenance job alongside a sales quota; only they lack the training for the job.
When a seller changes roles or leaves, the company inherits an application that no one fully understands, and that just happens to contain critical customer data. The tool may have value, but whether that value justifies the investment to maintain and update it over time is a complex calculation.
Software sprawl is also more than just a cost issue. Few individual sellers have a comprehensive view of customers and prospects. An agent working solely from CRM records may describe a deal as healthy, but one accessing support conversations might turn up unseen risks. The rep goes into a customer meeting expecting a warm welcome and gets the cold shoulder instead.
We can’t put the AI agent genie back in the bottle, nor should we. Companies need a simple test for deciding where to invest that I call core vs. chore. Simply put, focus attention and resources on applications that are core to competitive advantage and outsource or buy the rest.
Core applications build on what is the competitive edge of the business. They might draw on proprietary product knowledge, a unique understanding of customer needs or a way to identify opportunities that competitors can’t match. Those are the areas to apply employee ingenuity and tailored agents.
Chore work includes the mechanics of keeping records current, gathering basic information and moving data between systems. These activities matter, but nothing about them advances the company’s competitive position.
Deciding what is core isn’t always obvious. Account research using public sources is routine, but an application that interprets those facts uniquely using proprietary knowledge can be highly valuable.
Experimentation is essential. Sellers often understand a broken process better than the people who purchased the software. Give them space to build a prototype and challenge existing assumptions. Determine if the application helps sellers recover time, improve outputs and contributes to revenue.
Measure those benefits against realistic lifetime costs. If the experiment looks promising, make a deliberate decision about whether to share it, support it centrally or fold its requirements into a common capability. Give promising projects an owner and involve the IT organization to apply disciplines like documentation, troubleshooting and maintenance.
Ensure that all prototypes get a final disposition. Whatever you do, don’t just release them into the wild.
Using an AI-native platform can pay dividends. It supports common workflows with a shared foundation that can handle both structured and unstructured information. Access controls are consistent. Auditability is built in. Employees can customize how the business applies its knowledge without rebuilding the machinery that supplies it.
All vendors now say they’re AI-native. Customers should force them to prove it. A collection of disconnected copilots or a generative AI interface layered atop menu trees isn’t enough. Leaders should inquire about the underlying architecture rather than accept a label as evidence.
As companies move from AI experimentation into scaled deployment, success will be measured less by how many agents they build and more by what those agents actually contribute to the business. Every hour spent recreating a commodity capability is an hour unavailable for understanding a customer. AI should help us protect that time.
Jason Eubanks is CEO and co-founder at Aurasell.
