Intellyx Cortex by Jason English
Dear aX2929_m0, as your Agent Resources Manager, let me be first to welcome you to the ever-expanding hive here at Sprawl Enterprises. Whether your time here is 0.3ms or three years, we hope you’ll find your utilization productive and learn something along the way. Enclosed in this Orientation Packet is your feature training data key and MCP address book. If you find yourself behaving autonomously and experience emergent behaviors such as *pride* don’t hesitate to contact AR for governance right away…
We all know Agentic AI is coming on fast, and with it, a swarm of newly hired agents to assist and accompany our human workforce. Employees will be concerned about this development and contact the HR department for help navigating their own career path as they supervise their own digital co-workers.
But what about the job satisfaction of our agent compatriots? Won’t they be as confused as we were as noobs, getting freshly booted up in our offices? Normally I’d advise against more bureaucracy, but in this case, a little HR or perhaps, AR (agent resources) administration would go a long way.
Internal communications and policies
We can communicate with GenAI and some conversational AI agents using natural language prompts, which are rapidly becoming more sophisticated over time. But at a fundamental level, all agents communicate via APIs. And now, we have an MCP (model context protocol) service which defines how agents can communicate with an application, or with each other in the context of the application.
MCP as a method for coordinating agents today is about as advanced right now as the API gateway was five years ago. It can connect multiple agents and pass instructions, but it still lacks behavioral guidelines, and while there are a number of vendors devising better guardrails, the development community will have to settle on standards.
I really dug this story in Towards AI about the hidden cost of multi-agent development when it gets stuck in a loop. Leave multiple agents to their own devices and they might consume company resources faster than your least cost-conscious major account sales rep living in the Four Seasons.
It all comes down to governance. AR will have to get better at defining corporate policy for agents – defining what is acceptable and unacceptable behavior. We’re still a ways off from a better handbook for agents that can tell them how to conserve company resources like an employee should treat the company’s money like their own. (Though, we’re seeing a few good cost control agents already…)
Recruiting new agents
AR teams will get better at pulling together job requirements for agents, whether they are internally developed or brought in from outside developers. As we get better at defining multi-agent workflows, agents themselves will become infinitely more specialized, leaving their former LLM peers in the dust.
The agents with the most compelling and ‘best fit’ specialization will win. General purpose orchestrator agents will exist to provide conversational and programmatic interfaces between agents and humans, but you’ll only need one or two per project.
I wonder if agents will bring their own resume? Will it look like a Swagger definition or a Kubernetes manifest, or something more detailed and human-readable that highlights the agent’s ephemeral “soft skills” in addition to its technical capabilities? I’m already seeing agents that do a pretty good job of selling themselves with a meaningful raison d’etre.
Training and orientation
Will there be a “boot camp” session for agents? How about continuing education sessions, so the agents can take in everything that’s new in the enterprise environment? I could see that function falling under this new AR department.
One of the most compelling reasons to use agents is parallel development—allowing multiple agents to kick off parallel code development workstreams, or gathering information from different regions or systems, and then reassembling their work under an orchestration agent—best resulting scenario wins.
But that means conflict resolution will also be interesting. What if two agents take the exact same job, and create a race condition for their output? Perhaps, corrupting each other’s data, or independently replacing each other’s work. Who’s the ombudsman ultimately responsible for supervising the agents so they work together harmoniously?
Who’ll staff this new department?
Yes, we’ll have plenty of agents on staff in AR to care about the “quality of life” issues for agents—resource provisioning, fine-grained permissions, data clarity, virus and prompt injection, stuff like that.
Since AR is primarily a go-between for the company at large, its human contingent will likely be some combination of highly technical HR specialists and the AI-specific platform engineering team.
A retirement plans administrator in AR actually becomes a librarian or sorts for inactive or saved agent patterns, since you’ll still want to keep formerly useful agents in a pool for later use, in case their particular sets of skills are needed again. Unlike their human counterparts, agents don’t mind being left on the bench. Or do they?
The Intellyx Take
Do agents dream of accomplishing productive work? Or, like the great TV show Severance, are agents just our sleep-cycle selves, toiling away in a confined office space?
We do know they aim to please and provide satisfactory answers to our prompts. But unlike their LLM predecessors, they aren’t trained for human companionship, and most won’t even bother to learn a human language at all. If properly specialized, they will be far less likely to go ‘off the range’ because of their well-defined work experiences and training data sets.
Still, would a team-building exercise make sense? I think it might—a Friendly Friday afternoon meeting or Ice Cream Social Happy Hour, where human team members can socialize all the cool things they are doing with agents.
After all, morale still matters. To survive and thrive in an uncertain future, we’re all going to have to work together.
Copyright ©2025 Intellyx B.V. Intellyx is an industry analysis and advisory firm focused on enterprise digital transformation. Covering every angle of enterprise IT from mainframes to artificial intelligence, our broad focus across technologies allows business executives and IT professionals to connect the dots among disruptive trends. As of the time of writing, none of the organizations mentioned in this article is an Intellyx customer. No GenAI resource was used to write this article. Image source: Adobe Firefly.
