The New Operating System for Product Management
Building a product in the current landscape feels less like traditional engineering and more like conducting a chaotic orchestra of digital entities. We are moving away from the era where a Product Manager (PM) simply sat between the user and the developer. Now, the PM sits between a dozen different AI agents, each processing information at a speed that humans weren't built to match. The biggest challenge isn't the technology itself; it is the cognitive load of switching context every five minutes.
When you are managing multiple agents, you aren't just delegating tasks. You are managing state. If Agent A is working on the database schema and Agent B is drafting the UI, the friction occurs when the context from A doesn't flow naturally into B. For founders, this creates a bottleneck. If you spend all day syncing agents, you aren't actually building. You are just acting as an expensive manual bridge for data that should be automated.
The Multi-Agent Context Trap
Context switching has always been a productivity killer, but AI amplifies it. When you switch between human team members, there is a natural buffer—a greeting, a moment of recap, a shared understanding of the project's history. AI agents don't have that unless you provide it. Every time you switch your focus to a different agent, you have to reconstruct the prompt logic in your head.
The solution for builders isn't to work faster, but to build better silos. Successful teams are starting to realize that you shouldn't treat agents as generalists. You need specialized environments where the context is persistent. Instead of one massive thread where you ask a bot to do everything, you need discrete channels for logic, design, and copy. This reduces the mental tax on the human operator and keeps the output from devolving into a generic mess.
Giving Feedback to Non-Human Designers
One of the most frustrating parts of the current AI wave is reviewing generated designs. We have all been there: the AI produces something that looks 90% correct, but that final 10% feels uncanny or just plain wrong. The instinct for most PMs is to give vague feedback like "make it look more professional" or "clean up the layout."
AI doesn't understand "professional." It understands tokens and spatial relationships. To get better results, you have to stop talking like a manager and start talking like a grid system. Builders who are winning right now are those who provide feedback based on constraints. Tell the agent to adjust the white space by specific ratios or to align elements to a specific 8-pixel grid. The more you treat the AI as a precision tool rather than a creative partner, the less time you will spend in a loop of mediocre iterations.
The Fight Against the AI Accent
We are currently drowning in "AI-speak." You know the style: overly enthusiastic, peppered with words like "delve," "tapestry," and "comprehensive," and filled with bullet points that say nothing. It is the corporate jargon of the machine age. For founders, this is a branding nightmare. If your product sounds like a chatbot, users will treat it like a disposable commodity.
Getting AI to write like a human requires more than just a good prompt. It requires a rejection of the default. The default setting for large language models is to be agreeable and safe. To break this, you have to feed the model your own actual writing—warts and all. Give it your internal memos, your Slack messages, and your unfiltered rants. Tell the model to ignore the standard rules of business writing and focus on brevity and skepticism. If a sentence doesn't add value, it shouldn't be there. The best AI-generated content is the kind where the reader can't tell a machine was involved because the tone is too specific to be generic.
The Transparency Gap in Product Management
There is a recurring theme in the community about why PMs are hesitant to share what they are actually working on. It usually isn't because they are hiding secrets; it is because the work is increasingly messy. In the early stages of a build, things change every hour. Sharing a roadmap on Monday that becomes obsolete by Wednesday feels like a failure, even if it is just a natural part of the iteration process.
However, this lack of transparency creates a trust vacuum. Founders and engineers need to see the "ugly" version of the work. The shift toward AI-assisted building means the speed of iteration is faster than the speed of reporting. The only way to fix this is to move toward real-time documentation. Instead of polished weekly updates, builders should be sharing raw outputs and internal logs. It is better to be transparently pivoting than to be silently perfect.
The Founder's Takeaway
The transition to an AI-augmented workflow isn't about finding the perfect prompt; it is about managing the flow of information. If you find yourself overwhelmed by context switching, it is a sign that your processes are still designed for a pre-AI world. You need to harden your context, get technical with your design feedback, and ruthlessly strip the "AI accent" from your communications. Building in public isn't just a marketing strategy anymore—it is a survival mechanism for teams trying to keep up with the speed of their own tools.
Read the original at Lenny's Newsletter →