Sachin Kumar S. & Kishor Ravindranath Patil
Engineering Leaders
The AI revolution is becoming a victim of its own success.
As organizations become more comfortable using AI across their divisions, workflows, and products, they’re encountering a new problem: Enterprise AI specialists are in such demand that they’re too overwhelmed to handle all the pressing innovations and implementations across their organizations.
Any organization that wants AI to make an impact and deliver demonstrable results will have to figure out how to deliver it at scale. But the solution isn’t to hire more AI specialists. Instead, the way forward is to spread AI knowledge, resources, and mindset across all their engineering teams.
In other words, instead of bringing products to the AI team, bring AI to the engineering teams.
Every project is different, coming from different divisions with unique needs, data structures, restrictions, and business goals. Each time an engineering team brings in an AI team, the AI team must learn and understand the business context before they can perform their work.
That’s a tall order for any one project, let alone one after another. Not only does the AI team have to learn the business goals and context of each division and project, but they have to get into the nitty-gritty, time-consuming minutae. When they need to understand a dataset with 100 columns, it’s a challenge. When they have to do this for every single project, it’s a problem.
There’s nothing magical about AI. It’s just another set of tools like anything else, a set that any competent engineering can figure out how to deploy, especially when they know what they need to achieve and are familiar with the structure and meaning of their data. (It’s certainly much easier than trying to get the AI team to learn every domain, business need, and organizational subculture.)
For example: If you want to add the ability to place an order through an existing system, you don’t need to set up a whole new AI infrastructure. Instead, the hard part is giving context to an LLM. The existing team already knows the context. How you want to do that is up to you: vector search, database search, code, or other methods.
What’s needed is an easy, quick, and secure way to put these tools in the hands and rank-and-file engineering teams.
The first thing is to create the framework for engineering teams to deploy AI when and how they need it.
The ideal system should adhere to your organizations security requirements and restrictions, be well documented, and include some budgeting guidance so teams don’t have sticker shock when you send them the bill for their AI usage.
Don’t tie yourself to any one LLM. Instead, build a system that allows you to tap into all the available systems, as well as new ones as they arise. That way, teams can experiment and find the best LLM for each particular use case.
One promising technology is Model Context Protocol (MCP), developed by Anthropic, a platform-agnostic framework that allows teams to call whatever LLM they want to try. After all, they know the problems they’re trying to solve, so they’re in the best position to evaluate the responses.
Most enterprise application development involves asking, “How can I ask questions and get the right answer?” and then building chatbots to do the work. So, companies typically just copy content from various divisions to the AI division to let them build on it.
This copying creates serious scaling problems, like storage costs, security risks, and keeping data evergreen.
Rather than copying data to a central location, let teams work with their data where it already lives. This is where AI Agents can help, by understanding each individual dataset, and then passing along analyses to AI Superagents and individual product teams. This eliminates the copying problem while ensuring teams get real-time access to current information.
Once teams have access to AI tools, they can experiment with different approaches and find what works best for their specific challenges. Different departments may discover that different models excel at their particular use cases, and they can even get creative, such as using one LLM to evaluate the work of a competing LLM, free of biases.
Equally important is sharing information with other teams. This is where the AI team can be a particular resource, collecting use cases and making recommendations to teams. When everyone shares, everyone learns.
One of the biggest barriers to AI adoption isn’t technical, it’s psychological. Leaders worry about hallucinations, unexpected costs, and compliance issues when deploying AI solutions to production.
The solution is to create safe spaces for experimentation before anything goes live. Internal AI playgrounds — essentially enterprise-grade, private ChatGPT-style sandboxes — give teams a controlled environment to test ideas, catch problems, and build confidence within your private network.
These playgrounds serve multiple purposes. They allow teams to validate AI approaches, identify potential hallucinations or edge cases, and understand cost implications before committing to production. More importantly, they foster an AI-first culture by making the technology accessible for everyone, not just specialists.
When teams can quickly prototype an AI feature and get feedback from stakeholders (all within a secure environment) they move faster from idea to production-ready solution. This dramatically reduces the time and risk involved in AI deployment while building organizational confidence in the technology.
None of this eliminates the need for AI expertise; it simply changes the AI team’s role from building applications to building and maintaining infrastructure for others to use effectively. The other teams, meanwhile, can focus on understanding their divisions’ business needs and responding by identifying the most effective and relevant tools.
The technology exists, complete with proven frameworks. Now that AI is reaching a new level of maturity and adoption, it’s time to break through the enterprise bottleneck.
Sachin Kumar S. and Kishor Ravindranath Patil are Senior Engineers at Tricon Infotech, specializing in AI implementation and enterprise software architecture.
