Guest: Boris Renski
Company: Apelogic
Show: The Agentic Enterprise
Topic: Agentic AI

Companies are making the same mistake with AI that they made during the Dot-com boom: over-specialization. Instead of building strategic AI capabilities, they’re buying dozens of specialized agents—one for sales, one for support, one for data analysis—creating a fragmented, expensive mess that often makes problems worse. The smarter path? General purpose agents connected to enterprise data.

Boris Renski, CEO of Apelogic, has seen this pattern firsthand. His company was born from solving a data challenge at Helium Mobile, a US mobile plan operator with over 600,000 subscribers and 120,000 network access points generating massive amounts of data. Initially, they built what seemed like the obvious solution: a specialized data agent. It didn’t work as expected.

“We built a relatively heavy harness around that agent with all the details of what the database looks like, what the different business terms are, and how it should follow a particular flow to query the database,” Renski explains. “After playing with it for a while, we realized it didn’t actually work very well. It works okay, but all too often it doesn’t give correct answers. It takes a long time, and the whole thing is pretty brittle.”

The team pivoted to a radically different approach: stripped-down MCP connectors that let employees use Claude directly with their databases. The results were dramatically better. This experience shapes Renski’s broader thesis about enterprise AI adoption.

“There is pretty heavy over-investment in specialized agents and not enough emphasis on general-purpose ones and on making folks within a company effective at using general-purpose agents,” Renski says. General-purpose agents can already handle approximately 95% of what specialized agents do, and that percentage is climbing as frontier models improve.

The pattern mirrors the Dot-com era. Just as every category didn’t need its own specialized website (remember Pets.com?), not every business function needs its own specialized agent. Amazon consolidated e-commerce; general purpose agents will consolidate enterprise AI use cases.

What makes general purpose agents more effective? Renski points to the tight integration between how frontier labs train their large language models and build the harnesses around them. “Anthropic spent a lot of cycles training the latest set of its frontier LLMs to work and prefer and be biased towards CLI tools versus MCP,” he notes. This gives frontier labs using their own models an inherent advantage over companies wrapping external LLMs in custom harnesses.

For Helium Mobile, the impact was immediate and measurable. Before Apelogic, data queries required specialized SQL expertise concentrated in just three people. Mario Di Dio, GM of Networking at Helium Mobile, describes the old reality: “Each single different business function has different needs on how to slice and dice the same amount of data. Technical people like Joey wanted to see it in a certain way, there’s a BD person and sales people and marketing. It was very hard to fit all these needs in the superset dashboard.”

The result was data fragmentation and inconsistency. “Each single different SME had different similar SQL queries, so we would end up with different numbers many times,” Di Dio explains. Marketing, engineering, and finance teams each maintained their own databases with duplicated data structured differently for their needs—an expensive, inefficient approach that inflated AWS bills and created bottlenecks.

Joey Padden, VP of Network Architecture at Helium Mobile, saw the operational transformation: “We had probably three people who were in charge of SQL interaction. With the advent of the Apelogic product, we can have people on the support team, executives asking questions of the data, and I’m not worried about them not getting the question or context correct.”

The efficiency gains extended beyond just faster queries. Content teams could pull data for blog posts about network coverage during events like Mardi Gras or the Super Bowl without waiting for engineering support. Sales representatives could access customer data instantly. The cycle time for business insights collapsed.

But even infinitely intelligent general purpose agents need something: context. This is where Apelogic fits in. The platform provides what data scientists call a “semantic layer”—a memory system that helps Claude understand company-specific definitions and database schemas.

“You need to have context about the data that it’s querying,” Renski explains. “It needs to understand what the database schema looks like. It needs to understand things like what is Helium Mobile’s definition of monthly active user. Those things are very specific to a company.”

The semantic layer evolves as employees use it. When someone asks about Wi-Fi hotspots with bad performance, Claude might not initially know what “bad performance” means. Users can correct it, defining the term based on signal strength thresholds. Apelogic remembers that definition and suggests saving it so future queries use the same business logic.

This creates a virtuous cycle: more usage generates better context, which enables more accurate answers, which encourages more adoption. It also consolidates scattered tribal knowledge into a system that survives employee turnover.

Renski’s advice to C-level executives is blunt: “Stop buying specialized agents and start focusing on using general purpose ones and making the general purpose agents more effective within your organization.” Instead of purchasing yet another specialized agent for data science or customer support, companies should invest in connecting Claude or similar general purpose agents to their enterprise tools and data.

The architectural shift also promises cost savings. By eliminating duplicated databases across departments and consolidating queries through a single semantic layer, Helium Mobile expects measurable reductions in their AWS bills. More importantly, they’ve unlocked efficiency gains that translate to faster customer engagement, better data-driven decisions, and ultimately more revenue.

The broader market is starting to recognize this pattern. While peak “AI will kill SaaS” hype has dominated headlines, Renski predicts a more nuanced future. “I don’t think SaaS is going away,” he says. “The interface will have to change. It’s going to become dynamically composable.” The future likely combines chat interfaces with rich UX elements generated on demand—not pure command-line interactions or traditional dashboards, but something new that adapts to context.

For Apelogic, the long-term vision extends beyond data queries. “As software in general becomes easier to build, there are very few moats that remain,” Renski observes. “One is the data stored in enterprise databases, and the other is enterprise context.” Apelogic aims to become that context layer—the enterprise memory system that defines workflows, business terms, and organizational knowledge in an agentic world.

The lesson for enterprises is clear: the AI winners won’t be those who accumulate the most specialized tools. They’ll be the organizations that empower employees with general purpose agents deeply connected to company data and context. The specialized agent gold rush may be this era’s Pets.com moment. Don’t get caught buying the hype.