For the first year at OpenLoop, we hired the exact way every other software agency did. We looked for deep specialists. We wanted the best frontend developer we could find in Srinagar, the fastest backend engineer, and the most detail-oriented QA tester. It was the standard playbook for building a tech services company.
But by mid-2026, that playbook broke entirely. The tools got too good, and they got good too fast. We realized that hiring a person to execute one specific, narrow technical task was no longer a competitive advantage.
An AI agent could do it faster, cheaper, and often with fewer syntax errors. The bottleneck was no longer writing the code. The bottleneck was figuring out what code actually needed to be written to solve a real business problem.
We completely changed our hiring strategy across the board. We stopped looking for narrow specialists. Instead, we started hunting for a very specific, much rarer type of person. We started looking for the AI generalist.
The Death of the Narrow Workflow
When you build custom software for small businesses, you notice a pattern very quickly. The hardest problems are rarely technical anymore. They are connective and operational.
Our clients don't just need someone to write a Python script. They need someone to understand their entire daily operations pipeline, identify the actual points of friction, and string together multiple AI agents to solve them. It is entirely about orchestration, not just execution.
Writing boilerplate code is becoming a commodity. However, understanding how to apply AI agents in business remains a massive, un-commoditized skill. It requires a holistic view of the company.
A specialist looks at a client's problem and asks how their specific skill can solve it. If they are a React developer, every problem looks like a frontend web app. An AI generalist looks at a problem and asks a fundamentally different question. They ask which combination of digital and human workers is the most efficient way to solve this exact bottleneck.
Sometimes the answer is a web app. Most of the time, it is a well-prompted agent reading emails, extracting data, and pinging a WhatsApp group. The generalist doesn't care about the stack. They only care about the outcome.
This reality has completely reshaped our internal operations. We no longer spend weeks debating which specific framework to use for a minor internal tool. We spend that time refining our prompts, structuring our data, and ensuring our agents have the right permissions to do the heavy lifting.
Managing Digital Employees
The most important realization we had at OpenLoop is that AI agents are not just software tools. They are digital employees. And just like human employees, they require management, oversight, and a very clear set of instructions to function properly.
The people who thrive in our current environment are not the ones who want to keep their heads down in an IDE for eight hours a day. The people we actively want to hire are managers of digital labor.
They need to understand enough code to know when an agent is hallucinating a library that doesn't exist. They need to understand enough design to know when a generated UI is actively hostile to the user. Above all, they need enough raw business sense to know if the workflow actually saves the client money in the real world.
We are essentially hiring product managers who can deploy and direct AI agents as their engineering team. This is a profound shift in what we value in an interview. We no longer do whiteboard coding tests. We give candidates a messy business process and ask them how they would automate it using three different AI agents.
We want to see how they think about failure states, human hand-offs, and data flow. The actual implementation details can be handled by the models later. If a candidate cannot clearly articulate how to manage a rogue agent that goes off-script, they are not ready for a modern production environment.
Managing these digital workers requires a high degree of empathy and critical thinking. You have to anticipate where the AI will get confused and build guardrails to prevent catastrophic errors before they happen. That is not a specialized coding skill. That is a generalist management skill.
Get new posts by email, daily-ish writing on ISP economics, AI in production, and building in Kashmir.
The "All-Around Athlete" and Our Heritage
There is a specific term we use internally for this kind of person: the all-around athlete. When you are building companies in an emerging market like Kashmir, you don't have the luxury of massive, highly specialized departments.
At ViberNet, our fiber ISP, a field technician often has to understand basic routing protocols and customer psychology, not just how to splice glass. They need to be able to explain complex latency issues to a frustrated homeowner while simultaneously recalibrating a junction box.
At OpenLoop, a project manager might be writing system prompts for an autonomous research agent while also handling client billing and defining the scope of a new project. You wear many hats, not because it is fun, but because survival demands it.
The constraints of our geography force a kind of ruthless pragmatism. The Shah Mir dynasty historically thrived by being exceptionally adaptable and integrating diverse influences across the region to maintain stability. My own family name, Shahmiri, carries that exact legacy of adaptation. We cannot afford rigidity, neither in our history nor in our current business models.
This is exactly why the generalist is so incredibly valuable to us today. They are inherently adaptable. When the underlying technology stack completely changes every six months, a deep specialist suddenly becomes obsolete. A generalist simply learns the new abstraction layer and keeps moving forward without missing a beat.
The Future of Agency Work
The traditional agency model has always been about billing by the hour for specialized human labor. You pay for the senior developer's time, the junior developer's time, and the designer's time. It is a slow, expensive, and linear model that breaks down as soon as client demands scale up rapidly.
The new model we are building is entirely different. It is about billing for the business outcome, delivered by a fleet of AI agents that are orchestrated by a few exceptionally smart human generalists. The client doesn't care how many hours it took to build the tool. They care that the invoice processing time dropped from three days to four minutes.
This shift isn't just about cutting operational costs. It is fundamentally about speed and scale. A small team of AI generalists can handle the total workload of a traditional agency five times its size, provided they know how to build the right autonomous systems.
The leverage you get from one great generalist directing a swarm of specialized agents is absolutely staggering. It allows a small firm in Srinagar to compete on sheer output and quality with much larger firms anywhere in the rest of the world.
We are no longer selling hands on keyboards. We are selling a managed, scalable infrastructure of intelligence. And to manage that infrastructure, you need people who can see the entire board, not just a single square.
Adapt or Become Commoditized
The tech industry spent the last decade telling everyone to find a niche and specialize deeply. We were told that the riches were in the niches, and that being a generalist meant you were a master of none.
That was fantastic advice when humans had to do all the typing. It is terrible advice today.
If you are planning your career in tech right now, my advice is very simple and very blunt. Stop trying to be the absolute best at one tiny slice of the stack. The models will beat you at that game eventually, and they will do it without asking for a coffee break.
Instead, learn how the whole system fits together. Learn how to talk to clients, how to design a resilient architecture, and how to manage the autonomous agents that will inevitably do the specialized work. Learn how to translate messy human needs into structured data that an LLM can understand.
The future belongs to the orchestrators. The specialists will just be the APIs they call.
