AI Stack Series: 4 System-Based Shifts Redefining the Engineering Org

For our second installment of the AI Stack Series, we gathered Shopify’s Mikhail Parakhin, Factory’s Eno Reyes, and Meta, Google DeepMind alum Mat Velloso to discuss the engineering transformations happening now, from AI-assisted coding to autonomous orchestration

The era of AI-assisted coding, where tools like Copilot and Cursor primarily helped engineers generate code faster, is giving way to a new era of autonomous orchestration. As AI moves beyond the text editor to managing long-horizon missions and navigating operating systems, engineering leaders face a fundamental transformation in how technical organizations are built and managed. 

To explore this frontier, our recent AI Stack Series: Engineering session outlines a strategic roadmap for founders and CEOs shifting from individual efficiency to agent-native engineering systems.

Shift 1: Autocomplete to fully autonomous code generation

“We’re entering a post-editor era,” Shopify CTO Mikhail Parakhin says. Before leading Shopify’s frontier technology organization, Mikhail served as CEO of Microsoft Ads and CTO of Yandex, where he created foundational infrastructure such as ClickHouse.

At Shopify, the usage of traditional editor-first tools is declining in favor of systems that handle the sheer volume of AI-generated contributions. It’s a transition marked by an extreme concentration of productivity, and Mikhail notes that token consumption is becoming heavily weighted toward the top percentile of engineers. 

“Writing code is now incredibly archaic,” Mikhail says. “It’s like compilers. Once they came in, people stopped looking at assembly. Now, people are starting to not look at the code at all.” 

This shift is driven by a large surge in volume: Shopify saw pull request complexity increase by about 50% in just one year. To manage, Shopify has built custom systems, such as River, which allow engineers to deploy code without ever cloning a repository. This is significant because it shifts the engineer’s role from managing local environments and infrastructure to directing high-level product intent, while AI agents handle the mechanical plumbing of the codebase in the background.

According to Mikhail, Shopify has shifted from human-led reviews to what he calls “a narrow waist strategy” that entails high-reasoning reviews and critique loops. “We use the highest reasoning models to catch bugs that humans miss at high volumes,” he says, “and employ multiple agents from different providers to debate and critique each other” to provide an automated safety net. 

Automated safety nets are where a company like Namespace comes in. Built for this new way of working, the NEA-backed company provides the underlying muscle that lets AI agents skip standard code review. By giving agents their own dedicated computers to test in, Namespace builds validation right into the process, essentially making manual oversight a thing of the past. 

If the assistant era was about fixing a single bug, Factory CTO Eno Reyes says, the agent-native era is about missions. A leading expert in autonomous code generation, Eno has worked as an Machine Learning Engineer at Hugging Face and a Software Engineer at Microsoft, where he built production-scale large language models.

Engineers need a harness, he adds, not simply a nicer user interface. At Factory, harnesses manage the technical overhead — performance, tokens, and model-switching — and allow AI to autonomously work in the background.

Factory also uses droids, autonomous units that execute long-running tasks and can run unattended for up to 15 days. Unlike simple chat interfaces, the droids modernize entire codebases faithfully against high-level plans. It’s a shift in the engineering workflow from manual labor to high-level orchestration, Eno notes, where progress is measured by the execution of a multi-feature task list.

Shift 2: Single agents to the agent harness

If the assistant era was about fixing a single bug, Factory CTO Eno Reyes says, the agent-native era is about missions. A leading expert in autonomous code generation, Eno has worked as a machine learning engineer at Hugging Face and a software engineer at Microsoft, where he built production-scale large language models. 

Engineers need a harness, he adds, not simply a nicer user interface. At Factory, harnesses manage the technical overhead — performance, tokens, and model-switching — and allow AI to autonomously work in the background. 

Factory also uses droids, autonomous units that execute long-running tasks and can run unattended for up to 15 days. Unlike simple chat interfaces, the droids modernize entire codebases faithfully against high-level plans. It’s a shift in the engineering workflow from manual labor to high-level orchestration, Eno notes, where progress is measured by the execution of a multi-feature task list. 

Shift 3: AI is now moving to the operating system

The frontier of AI, Mat Velloso says, is moving beyond the text editor and into the operating system (OS) itself. With more than a decade of leadership experience, Mat has directed developer platforms and AI product strategies at Meta Superintelligence, Google DeepMind, and Microsoft. 

Agents are now capable of navigating the entire computer, Mat says, by using a screen and mouse to interact with tools beyond the codebase. This level of access enables autonomy that transcends code-only agents to complete tasks like parsing images of calendars and spreadsheets directly into functional data. “If we force users to type prompts, we’ve already failed,” Mat says. “The goal is to make AI interact with the computer as naturally as a human does.” 

Giving autonomous agents OS-level access does introduce significant security risks. “This shift requires a defense strategy against supply chain attacks and unauthorized access,” Mikhail says. What’s more, because agents make it so easy to automatically install packages, one single prompt could lead to immediate data exfiltration or a compromised system. 

The theoretical promise of models like Mythos poses an additional challenge. Because Mythos can read and interact with every layer of the OS, it has the potential to uncover long-dormant bugs in legacy systems that entirely predate modern security practices — exposing vulnerabilities that defenders didn’t know to look for. At NEA, we believe this friction is a major investment opportunity and a chance to rethink cybersecurity and secure tool use from the ground up. 

Shift 4: The talent profile redefined as CEO-engineer

The most significant shift may just be the changing definition of what constitutes a high-value engineer. The three speakers agree that an engineer who excels at manual syntax is a legacy profile. It’s now about an AI-native engineer, a high-leverage operator who treats their work with the high urgency and ownership mindset of a CEO. 

Since AI now handles the mechanical aspects of coding, the engineer’s primary value lies in deep product ownership and defining the intent of the code. The best talent will write precise specs and validate criteria to ensure that AI’s output actually moves the business needle. As Eno describes it, the goal is to move past simple prompts and instead use deep engineering intent to provide the AI with enough context and technical rigor that the resulting product is robust, scalable, and purposeful.

For a technical lead, the ability to manage an autonomous factory of agents becomes the ultimate competitive advantage in this automated world.

Success in autonomous AI requires leaning in

For founders, the goal is no longer helping engineers type faster, but empowering them to direct autonomous code factories. To succeed in this new era, teams need to move past simple chat boxes and start using specialized systems that can handle complex projects from start to finish, and which can also operate in pre-existing infrastructure systems like another engineer.

Simply put, engineers are now the architects who define the vision, while the AI navigates the technical implementation. The teams that win will be those eager to rethink their daily habits and embrace a future where AI does the heavy lifting.