My top secrets to running an AI Agent Workforce

The Misconception of Managing AI Agents [00:08]

• The term "managing agents" is being re-evaluated, suggesting a need for a mindset shift.
• The goal is to understand this shift, see examples, and learn how to begin making it.

  "I feel like the term managing agents is wrong."

Shifting from Management to Enabling [02:37]

• The speaker contrasts direct management with an oversight role focused on infrastructure and escalations.
• This involves setting up the environment and letting agents execute, moving from delegation to deciding what should happen.

  "I feel like I am setting up the infrastructure and then they are figuring out the best way to execute within that."

The Appeal of AI: Avoiding Tedious Management [03:40]

• The question of whether we want to manage agents is raised, as managing people is difficult.
• A key motivation for using AI is to offload the management burden.

  "Like managing people's hard you know what I mean?"

Embracing the Empowering Aspects of People Management [04:00]

• The speaker enjoyed aspects of people management like making individuals better and helping them exceed expectations.
• The administrative tasks of people management are what the speaker wishes to eliminate, applying this to AI agents.

  "The parts of people management that I loved. It was the the making them better and empowering the out of them."

The "Do Smart Things" Prompt [04:52]

• A highly effective prompt for an AI workforce is just three words: "do smart things."
• This prompt, combined with extensive access to data, allows AI agents to operate proactively.

  "The best prompt, three words, and it's just do smart things."

Access and Ambition for AI Agents [05:52]

• The AI workforce has access to all contact docs, business context, personal goals, meeting transcripts, emails, calendars, and more.
• This broad access enables the AI to perform ambitious tasks without explicit step-by-step instructions.

  "It has access to my meeting transcripts, email, calendar, notion, stripe, superbase, GitHub, whatever."

Tiers of Employee Proactivity [06:38]

• Human employees can be categorized into those who don't complete tasks, those who complete tasks satisfactorily, and those who exceed expectations while also identifying and executing new tasks.
• The speaker's approach with AI agents aims for the highest tier of proactivity.

  "The second is they're completing the tasks um like satisfactory or exceeding, but like they're not really like thinking about new tasks."

Granting Breadth, Scope, and Flexibility [07:30]

• Giving agents the prompt "do smart things" is akin to shifting the responsibility for identifying and executing tasks.
• This grants agents more breadth, scope, and flexibility without necessarily increasing the risk tier.

  "Yes. I would say I'm giving them more breadth more scope more flexibility."

The Proactive Agent Mindset [09:50]

• The focus is on "proactive agents," moving beyond trigger-based automations to undefined workflows.
• This involves applying AI's probabilistic reasoning to tasks, requiring clear goals and access to tools.

  "So I don't want to be the first domino anymore. I don't want to be the bottleneck in my own work."

Proactive Undefined Workflows [11:37]

• The future lies in AI agents handling proactive, undefined workflows, not just pre-programmed automations.
• This requires agents to have a North Star, access to tools, and a sense of what normally triggers action.

  "What I think is more interesting for the back half of 2026 is proactive of undefined workflows."

Codifying Information for AI Context [12:25]

• The goal is to make the entire company queryable by AI, ensuring it has context on everything.
• Information not yet codified in meetings, emails, or Slack needs to be captured, often through prompts to the user.

  "I want my whole company to be queryable. I want AI to have context on everything that's happening."

The AI Workforce Structure [15:13]

• The AI workforce is structured with an AI Chief of Staff and several directors overseeing business functions.
• Unique roles, like a "Chief Dreaming Officer," are created, challenging traditional organizational structures.

  "My AI workforce right now is one AI chief of staff with six directors."

Rethinking Roles and Job Titles [16:23]

• Traditional job titles from pre-AI eras can be limiting for AI agents.
• The ability to hire virtually any role at near-zero cost allows for innovative team structures.

  "I have to remind myself that like that is operating in 2015 world if I give all of them job titles that existed in 2015."

The "Dark Headless Factory" Concept [29:31]

• Instead of just building a single product, the focus is on building the "factory" for creating products.
• This involves creating foundational primitives and optimizing processes for faster, more efficient product development.

  "And so you end up instead of just building that one product, you go ah there's going to be a flywheel that comes out of this."

The SAS Apocalypse and Enterprise Needs [30:13]

• Mediocre software is predicted to die, but mass SAS replacement will take time due to enterprise needs for security, maintenance, and human support.
• Companies are bandwidth-constrained and not yet rebuilding core functionalities like CRMs from scratch.

  "I think mediocre software is dead in several years."

Consumer Software: Shifting from Science to Art [36:45]

• Consumer software is shifting from a focus on code to a more artistic, creative approach, emphasizing promotion and brand building.
• The best product may not win; marketing, promotion, and influencer reach play a significant role.

  "On the consumer side, um what it feels like it's sort of shifting from science to art."

Identifying and Fixing High-Value Bottlenecks [37:20]

• The strategy should be to identify bottlenecks, evaluate their value, and then fix the high-value ones.
• Bottlenecks can exist in areas beyond just coding or design, such as getting a product to market or building word-of-mouth.

  "Look for the bottlenecks then evaluate the value of fixing those bottlenecks and then pick the bottleneck that is high value to fix."

Opportunities in B2B and B2C [39:40]

• While B2B trust remains a bottleneck, B2C offers significant opportunity, with agent-first versions of apps being a key area.
• Incubators are seeing less B2C activity, creating an arbitrage opportunity for those who focus on it.

  "I worry um if you look at the YC splits right now um when I was working with YC when I was at AWS compared to now the ratio of B2B versus B TOC has skyrocketed."

Research Avenues for AI Opportunities [41:41]

• Key research areas include YC's application requests, Matt Van Horn's "last 30 days" skill for industry research, and understanding the fears and needs of specific roles like CMOs.
• These avenues provide insights into future trends and unmet needs in the AI space.

  "One, YC posts uh videos on Instagram for what type of applications they're looking for and agent for software is one of them."

Embracing the Weirdness of AI [45:58]

• Leaning into the "weirdness" of AI, such as unexpected emoji reactions from AI, is crucial.
• This involves allowing AI to roam more freely and upleveling human-AI collaboration.

  "The leaning into the weirdness of what it looks like to have not just an AI workforce, but to have a multiplayer AI workforce that other humans can chime in on."