Microsoft is building an “agent factory,” and it might represent the biggest shift in how software gets made since 1975. While much of the industry fixates on how much code AI generates, Jay Parikh, EVP of Core AI at Microsoft, has a different vision: fundamentally rewiring how companies build intelligent capabilities.
But here’s what should keep every founder awake at night: “In two years, the companies that have figured this out will look entirely different,” Jay predicted during our conversation at the 2025 IA Summit. Companies that embrace this transformation and learn to build their own agent factories will lead, while others will find themselves “paddling faster” as their competition races away “in a speedboat.”
The insight that emerged from our conversation is profound: we’re not just automating coding, we’re entering what Jay calls “a completely different alternate universe” where software development itself gets reinvented. His agent factory concept — where foundation models, research breakthroughs, and production systems come together on an assembly line to produce AI agents — represents Microsoft’s bet on this transformation.
“For decades we’ve been building software where we go interview our customers, we come back, we design some schema, we write some business logic, and then we put some UI around it,” Jay explained. “Now, though, it’s not even that the paradigm goes on its head. It’s like a completely different alternate universe because now you have these models that can think, they can plan, they can reason, they can call different tools, they can collaborate with each other.”
The Factory That Bill Never Imagined
The concept crystallized during Jay’s first one-on-one meeting with Bill Gates. When Jay started describing his “agent factory” vision, Gates stopped him mid-sentence to share that when he started the company back in 1975, he always had this vision of a software factory.
Fifty years later, the parallel is striking but fundamentally different. Where Gates envisioned systematically producing software, Jay’s agent factory produces intelligence — coding agents, biology agents, fraud detection agents, code review agents. The raw materials aren’t just code and components; they’re foundation models, Microsoft Research breakthroughs, and reasoning capabilities that get assembled into production AI systems.
“I drew it out in ASCII Art for the team,” Jay said. “There’s these things I envisioned showing up on the loading dock — models, technology from MSR [Microsoft Research]. We need to put those together into a production line. Then, at the end, it produces some AI capability.”
This isn’t just about Microsoft’s internal transformation. Every company that wasn’t founded today will need to create a similar system, Jay argues. The agent factory represents “a cultural thing, a technology thing, an incentive thing, a systems thing” that organizations must embrace to remain competitive.
Beyond Productivity Theater
While industry leaders throw around impressive statistics — Anthropic’s Dario Amodei predicting 90% of code will be AI-generated by year-end — Jay takes a different view. Having scaled some of the largest engineering teams at Meta and Microsoft, he’s focused on capability, not code volume.
“These numbers around how much code is written by AI is largely uninteresting to me,” he said. “It’s really more about the capabilities it’s creating or giving to builders in the company.”
The real opportunity lies in eliminating what Jay calls “toil” — technical debt, framework upgrades, security fixes, performance optimization. “There’s stuff there that is just holding these very talented people back from achieving higher levels of creativity and collaboration. I want to shift that dramatically.”
His macro perspective is even more striking: We’re probably sub-1% of all software that’s ever been written in humankind compared to what’s going to be written in the next 10 years.
“We have this superpower that’s getting better and better every day in terms of being able to build prototypes and solve these types of problems,” he said.
The Evaluation Crisis
Despite Microsoft’s $80 billion investment in data centers and a comprehensive AI platform, Jay says there are still two critical gaps that resonate across the industry. First: evaluation systems that actually work in the real world.
“You can have a set of evals, build your application, evaluate some models, and be like, ‘Hey, we got these scores,’” he explained. “Then you put it in the hands of 10 customers and they vomit all over the experience.”
The problem isn’t technical benchmarks. Microsoft and others have gotten good at “one-dimensional evals.” He says there is a human touch, or four-dimensionality, to the lived experience of these products. Human experiences are personal, contextual, and resistant to standardized measurement.
The second critical gap: observability for enterprise AI. “Nothing in AI is going to work in the enterprise without observability,” Jay said. This includes monitoring, traceability, compliance, audit capabilities, and security integration. These are all areas where Jay expects “a vibrant ecosystem of startups” to emerge.
Build vs. Buy vs. Partner
Microsoft’s strategy reflects the fundamental tension every AI company faces: build proprietary capabilities or embrace ecosystem diversity. Jay’s approach is deliberately multi-pronged, revealing the complexity of winning AI strategies.
The Microsoft-OpenAI partnership remains “incredible for both companies,” he said, with collaboration across every level. But the platform strategy extends far beyond that exclusive relationship. GitHub Copilot offers models from multiple providers — Gemini, OpenAI, Anthropic, and Microsoft’s own models. Azure AI Foundry provides “thousands of models” for developers to choose from, he said. “And we’ll continue to invest in that breadth of choice.”
Perhaps most revealing: Microsoft has quietly released its own frontier models — the MAI family trained entirely in-house, with more coming. “There’s models that are Microsoft-trained that sit inside GitHub Copilot that most people don’t know about today,” Jay revealed.
The philosophy: “Being dogmatic about having only one choice or one model — that’s not what people want,” Jay explained. For founders, this reveals a critical insight: the companies winning the AI race aren’t picking one model or one approach. They’re building systems flexible enough to leverage multiple capabilities as the landscape evolves.
The Great Flattening
The transformation Jay envisions goes beyond technology to organizational structure. “Applications will collapse. I mean, merge, not collapse and go to zero, but I do think functions will collapse. And I think that’s going to be a big reckoning,” he predicted.
Companies that embrace flattened structures, combined functions, and shift time from “run the business” activities to creative work will “start to really lead.” Those that don’t? “There’s going to be a lot of enterprise teams that look like they look today, but where those companies are going — probably nowhere.”
The transformation goes beyond organizational charts. Current enterprise processes, where “somebody has to do 45 approvals before it gets into production at a big bank,” need fundamental rethinking. “Those systems, those functions inside of the company; we have to rethink them from the ground up.”
The Reasoning Revolution
Jay’s vision extends Microsoft’s 50-year journey from software factory to agent factory. Where the original Microsoft systematically produced software, today’s Microsoft is systematically producing reasoning capabilities that can be deployed across every business function.
The parallel to Gates’ original vision isn’t coincidental. It reflects Microsoft’s understanding that we’re not just in another technology cycle, but in a fundamental shift in how intelligent systems get built and deployed. The agent factory represents Microsoft’s bet that the future belongs to organizations that can systematically produce and deploy AI capabilities as efficiently as they once produced software.
From our conversation, it’s clear that Jay sees this transformation as both inevitable and urgent. The companies that build their own agent factories, whether using Microsoft’s platform or creating their own, will define the next era of technology. Those that don’t will find themselves stuck in productivity theater while their competition races ahead in speedboats.