From Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Last

The transition from software engineer to agent manager is no longer a distant future; it is the current reality for those building at the frontier of technology. Simon Last, co-founder of Notion, has been at the center of this shift, steering the platform from a human-centric collaborative workspace to a sophisticated environment where humans and autonomous agents work in tandem. By embracing AI as a core architectural principle rather than a mere feature set, Notion is redefining what it means to build, organize, and execute complex knowledge work.

Key Takeaways

• The Agent Manager Shift: Engineering has evolved from typing code to architecting, verifying, and managing autonomous agent workflows.
• Iterative Architecture: To keep pace with AI capabilities, developers must be prepared to rebuild their system harnesses frequently—often every six months.
• Democratizing AI: Success in the agentic era requires providing intuitive tools that allow non-technical users to build their own custom, autonomous workflows.
• Standardization for Agents: Just as APIs were built for human developers, platforms must now develop "agent-native" interfaces to ensure efficiency and reduce token costs.

From Writing Code to Orchestrating Agents

For years, the professional identity of a software engineer was defined by the act of typing code. Today, that definition is being aggressively rewritten. Last explains that his own daily workflow has undergone a total transformation; he rarely types code himself anymore. Instead, he spends his time designing end-to-end tasks, defining verification loops, and monitoring the output of his agentic swarm.

This shift requires a change in mindset from execution to supervision. When an agent is responsible for implementation, the human’s value shifts toward architecture and quality assurance. The goal is no longer to just "vibe code"—a term for haphazardly prompting an AI—but to think deeply about how to verify the agent's work, ensuring that it remains robust, tested, and aligned with system goals.

"I think of coding agents as the kernel of AGI. AGI will be a coding agent. Code is just a really, really useful primitive for representing deterministic logic." — Simon Last

The Philosophy of Constant Rebuilding

One of the most counterintuitive aspects of building for the AI age is the need for rapid, cyclical innovation. Notion has made it a practice to rewrite their AI harness roughly every six months. In a landscape where the underlying models and their capabilities are shifting constantly, clinging to a legacy system is a recipe for technical debt.

Designing for the New "User"

Historically, software platforms were designed for human interaction. Notion realized early on that an agent is a fundamentally different kind of user. Standard JSON structures for blocks, which are readable and efficient enough for traditional API development, can be prohibitively verbose and expensive for AI agents. By developing custom, agent-friendly markdowns and optimized SQLite interfaces, Notion has paved a way for agents to interact with databases and documents with minimal friction.

Building a "Switzerland for Models"

In a market saturated with powerful proprietary labs and emerging open-source giants, companies often face the fear of model lock-in. Last views Notion’s role as the "Switzerland for models"—a neutral ground where users can swap between the latest frontier models and high-quality, cost-effective open-source alternatives.

This flexibility is essential. Not every task requires the most expensive, massive frontier model. By providing access to a variety of tools, Notion ensures that the platform remains adaptable to the specific needs of its users. Whether a user is triaging emails with a custom agent or building a complex, autonomous feedback-routing system, the platform provides the infrastructure while letting the user choose the "brain" behind the logic.

Democratizing Custom Workflows

The true power of the agentic era lies in accessibility. When complex automation is restricted to senior engineers, the impact remains limited. Notion’s approach has been to push these capabilities into the hands of non-technical teams, such as their "People Team," who have become some of the highest adopters of custom agents. By providing a clear interface for agents to learn, store memory, and interact with tools like Slack and email, the barrier to automation is lowered.

"Once you get past the technical barrier of what is a prompt and how does an agent get triggered, I think it’s actually a very human-like interface." — Simon Last

This "memory" pattern—where an agent reads a blank page, reviews data, asks for clarification, and then creates a set of governing rules—allows users to build sophisticated workflows without writing a single line of code. It transforms the user from an observer into a designer of their own automated assistants.

The Future of Knowledge Work

As we look ahead, the definition of a "tool for thought" is clearly diverging from its origins. The platform of the future is not just a place to store information, but an active participant in the creative and operational process. Notion’s journey reflects this: the core primitives—documents, databases, and boards—remain vital, but they now serve a dual purpose. They are the artifacts of human creativity and the operating environment for the agents that accelerate our work.

The transition to agentic engineering is as much about cultural shift as it is about technical prowess. It demands an appetite for chaos, a commitment to empirical evaluation, and a willingness to stop doing the work ourselves to make room for the agents that can do it for us. As Last’s own experience shows, the measure of success in the near future will not be how many lines of code we write, but how effectively we manage the agents that write them for us.