How 80,000 companies build with AI: products as organisms, the death of org charts, and why agents will outnumber employees by 2026 | Asha Sharma (CVP of AI Platform at Microsoft)

Aug 28, 2025 Episode Page ↗
Overview

Asha Sharma, Microsoft's Chief VP of Product for AI Platform, discusses the shift to "product as organism," the rise of an "agentic society," and how companies and builders must adapt to the AI era. She shares insights on planning with "seasons," the importance of post-training, and leadership lessons from Satya Nadella.

At a Glance
11 Insights
57m 11s Duration
14 Topics
10 Concepts

Deep Dive Analysis

Shifting from Product as Artifact to Product as Organism

The Rise of Post-Training and AI Product Development

Patterns of Successful AI Companies and Common Pitfalls

The Renaissance of Full-Stack Builders

The 'Loop, Not the Lane' Organizing Principle

From GUIs to Code-Native Interfaces

The Rise of the Agentic Society

From Org Charts to Work Charts

Microsoft's Internal Use of AI Agents

Planning and Strategy in the Rapidly Changing AI Landscape

The Enduring Importance of Platform Fundamentals

Personal Motivations and the Impact of AI

The Future of AI: Reinforcement Learning and Post-Training Investment

Lightning Round and Final Reflections

Product as Organism

This concept describes products that are not static artifacts but living entities that continuously improve. They ingest data, digest reward models, and create outcomes, getting better with more user interactions, effectively 'thinking, living, and learning'.

Post-training

This refers to the process of optimizing models on a continuous loop, often using proprietary or synthetically generated data, rewards design, and rigorous A/B testing. It becomes more economically sensible for models beyond a certain parameter size, allowing for tuning to specific outcomes like price, performance, or quality.

Full-Stack Builder (Polymath)

In the AI era, a full-stack builder is an individual capable of handling multiple aspects of product development, blurring traditional functional lines like security, user research, and coding. This polymathic approach increases velocity and throughput, enabling faster iteration and metabolism of product changes.

Loop, Not the Lane

This is a new organizing principle emphasizing continuous feedback and product evolution over rigid functional lanes. It means that regardless of one's specific function, the focus should be on understanding and optimizing the end-to-end efficiency, cost, system design, and UI/UX of the product, with observability becoming a core cultural aspect.

Code-Native Interfaces

This describes a shift in product interfaces where direct interaction with code or text streams becomes more prevalent, moving away from traditional Graphical User Interfaces (GUIs). This trend is driven by the way text streams connect better with Large Language Models (LLMs), enabling greater composability and accelerating development, similar to historical shifts from desktop databases to SQL.

Agentic Society

A future state where the marginal cost of good output approaches zero, leading to an exponential demand for productivity that is scaled by AI agents. This society will see a proliferation of both embedded and embodied agents, fundamentally transforming work structures and potentially replacing traditional organizational hierarchies with 'work charts' focused on tasks.

Work Chart

In an agentic society, the traditional 'org chart' (hierarchy) transforms into a 'work chart,' where tasks and throughput are prioritized over reporting lines. This structure focuses on outward, task-based opportunities, with capable agents and empowered people collaboratively addressing issues, potentially leading to fewer organizational layers.

Seasons Planning Framework

A flexible strategic planning approach where 'seasons' are defined by significant secular changes in the industry or customer needs, such as 'the advent of agents.' This framework grounds everyone on the ethos, North Star metrics, and winning conditions, complemented by loose quarterly OKRs and 4-6 week squad goals, while deliberately leaving slack for continuous adaptation and disruption.

Invisible Work (Platform Fundamentals)

These are the often-undervalued underlying infrastructure and non-pixel-based aspects of a product or platform that are critical for long-term success. Examples include data residency, availability, reliability, privacy, safety, performance, and the right selection of tools, which are more impactful than numerous features.

Reinforcement Learning (RL)

A product technique expected to become highly important, especially in the 'post-training' phase, involving optimizing models through continuous feedback loops to achieve specific outcomes like price, performance, or quality. It allows for adapting off-the-shelf models rather than solely investing in building new ones from scratch.

?
What is the difference between pre-training and post-training in AI?

Pre-training involves massive initial compute and scientific expertise to create a foundation model, while post-training focuses on fine-tuning and optimizing an existing model with specific data and reward designs to achieve desired outcomes like price, performance, or quality.

?
Why is post-training becoming more important than pre-training?

Beyond a certain model size (e.g., 30 billion parameters), the capital expenditure for pre-training becomes economically less sensible. Post-training offers more leverage for economic efficiency and tailoring a model's behavior to specific use cases or desired outcomes.

?
What are common reasons AI companies fail?

Companies often fail by doing AI for AI's sake, launching many projects without a clear blueprint, or not treating AI as a real investment with proper measurement, observability, and evaluations. They also struggle with choosing the right tools from the vast and rapidly changing AI landscape.

?
Are traditional Graphical User Interfaces (GUIs) being replaced by code-native interfaces in the AI era?

Yes, a historical pattern shows interfaces moving from desktop to code (e.g., databases to SQL, cloud to Terraform). AI accelerates this trend, as text streams connect better with LLMs, favoring composability over canvas-based GUIs and requiring product makers to adapt their mindset.

?
How will AI agents change organizational structures?

AI agents will transform the traditional 'org chart' into a 'work chart,' where tasks and throughput are prioritized over hierarchy. Organizations will focus on outward, task-based opportunities, potentially leading to fewer management layers as agents and empowered individuals handle more responsibilities.

?
How do large organizations plan roadmaps and strategy in the fast-changing AI landscape?

Microsoft uses a 'seasons' planning framework, where a season is defined by secular changes (e.g., the advent of agents). This provides a shared ethos and North Star, complemented by loose quarterly OKRs and 4-6 week squad goals, with deliberate slack left in the system for continuous adaptation and disruption.

?
What is the current 'season' in the AI landscape, according to Asha Sharma?

The current 'season' is 'the rise of agents,' focusing on building blocks for agents to endure, including critical aspects like memory and tool-calling loops, to make them more complete and robust.

?
What are the most critical, often undervalued, aspects for a successful platform or product?

The 'invisible work' or platform fundamentals are most critical, including data residency, availability, reliability, privacy, safety, performance, and the right selection of tools. These underlying infrastructures, rather than just features or pixels, are what truly drive success.

?
What is the significance of Reinforcement Learning (RL) in the future of AI product development?

RL is expected to become one of the most important product techniques, especially in post-training, driving as much or more investment than pre-training. It allows companies to adapt and optimize off-the-shelf models for specific outcomes like price, performance, or quality, offering greater economic leverage and customization.

?
What is a key leadership lesson Asha Sharma learned from Satya Nadella?

Asha learned that 'optimism is a renewable resource.' Satya's ability to generate energy and clarity, using his optimism to renew everyone's dedication to the mission, is a crucial part of Microsoft's culture.

?
What is Asha Sharma's personal life motto?

Her worldview is about maximizing 'option value' rather than minimizing regret. This means prioritizing family, health, trust, and relationships, as these compound in the future and enable more adventures, rather than just avoiding past regrets.

1. Embrace Product as Organism

Shift your product mindset from static artifacts to living organisms that continuously improve through data ingestion, reward models, and user interactions. This approach, where products “think, live, and learn,” is the new IP for companies.

2. Prioritize Post-Training & RL

Focus investment on post-training, fine-tuning, and reinforcement learning (RL) rather than solely pre-training large models. This offers greater economic leverage and allows you to tailor models to specific outcomes like price, performance, or quality using your own, synthetic, or purchased data.

3. Cultivate Full-Stack Builders

Develop “polymath” or “full-stack” builders who are obsessed with the entire product loop, from efficiency and system design to UI/UX, rather than narrow functional lanes. This increases velocity and throughput, enabling faster iteration and adaptation in the AI era.

4. Become AI Fluent & Apply AI

Ensure everyone in your organization embraces AI and becomes “AI fluent” by using copilots and AI in daily workflows to raise skill ceilings and lower floors. Then, identify existing processes (e.g., customer support, fraud detection) and apply AI to them to achieve measurable impact and P&L benefits.

5. Build for the Slope

Recognize that technology is constantly changing, so bet on flexible platforms or app server layers that allow you to swap technologies in and out. This approach helps you adapt to continuous disruption rather than being beholden to any single tool.

6. Shift Mindset to Composability

Rewire your product-making mindset to prioritize composability over graphical user interfaces (GUIs), understanding how agents will read and interact with components. Focus on how products compose, scale infinitely, and facilitate collaboration, as text streams connect better with LLMs.

7. Plan with Seasons & Slack

Adopt a flexible planning approach by defining “seasons” based on secular industry or customer changes, aligning everyone on the North Star metric for that season. Set loose quarterly OKRs and 4-6 week squad goals, while leaving slack in the system for continuous disruption and new investments.

8. Focus on Invisible Platform Work

Understand that the true value and competitive advantage of platforms and products often lie in “invisible work” like data residency, availability, reliability, privacy, safety, and the right selection of tools, rather than just visible features. Prioritize these foundational elements for durability.

9. Foster Optimism & Clarity

As a leader, cultivate optimism as a renewable resource to generate energy and renew your team’s dedication to the mission. Provide clear direction on goals and leverage optimism to maintain commitment in a competitive environment.

10. Maximize Option Value in Life

Adopt a worldview focused on maximizing “option value” by prioritizing family, health, trust, relationships, and continuous learning. These areas compound over time, leading to more future adventures and a richer life, rather than just minimizing regrets.

11. Develop Mental Fortitude

Engage in practices like Taekwondo (or similar disciplines) to cultivate mental clarity, courage, ambition, and unwavering resolve. This mental training is crucial for navigating challenges in both work and life, alongside learning to meditate and clear your head.

All of a sudden, these are these living organisms that just get better with the more interactions that happen. And in many ways, I think this is the new IP of every single company.

Asha Sharma

I think it's all about the loop, not the lane here. And so I think that whatever function you are, you have to be obsessed with trying to understand like the efficiency or the cost of the product, the actual rewards or like, you know, system design that you're going after, the actual UI, UX, how that actually manifests for agents or people.

Asha Sharma

We're approaching this world in which the marginal cost of the good output is approaching zero. We're going to see exponential demand for productivity and outputs. The way that you scale to that is with agents.

Asha Sharma

One of my favorite ones, so right now, are my engineering partners out, so I jump on the live site bridges when something goes down and, you know, as something as simple as, like, you can automatically get a summary of everything that just happened because usually there's 15 people talking, you don't actually know where the incident started, where it's going to end and everything, and then all of a sudden I have that and I can kind of figure out and ask questions and get updates. It's, like, awesome.

Asha Sharma

Optimism is a renewable resource.

Asha Sharma (attributing Satya Nadella)

What AI will help us do from a workforce perspective? What it will help us do from a healthcare perspective? Like, you know, my mom has cancer and I think a lot about how, wow, we might find a way to solve the form of cancer she has in my lifetime and I never thought that was possible three years ago. Like all of that's deeply profound.

Asha Sharma

Three-Phase Pattern for Successful AI Companies

Asha Sharma
  1. Embrace AI: Ensure everyone in the organization becomes AI-fluent, using copilots or AI in daily workflows to understand its potential and raise the ceiling while lowering the floor for skills and tasks.
  2. Apply AI to existing processes: Identify an existing process (e.g., customer support, fraud detection) and apply AI to improve it, mapping out the process, seeing the impact, and realizing P&L or intrinsic benefits.
  3. Use AI to inflect growth: Leverage AI to improve customer experience (boosting LTV or retention), co-create new concepts or categories, or evolve embedded agents to embodied agents capable of handling an exponential number of tasks.

Microsoft's 'Seasons' Planning Framework

Asha Sharma
  1. Define the 'Season': Ground everyone on the ethos, secular changes happening in the industry or from customers, and what winning looks like, setting a North Star metric. A season can last a year, six months, or three months (e.g., 'the advent of agents').
  2. Set Loose Quarterly OKRs: Based on the current season's direction, establish objectives and key results for the next quarter.
  3. Establish Squad Goals: Teams operate in squads, setting 4-6 week goals for specific problem areas that ladder up to the quarterly OKRs.
  4. Leave Slack in the System: Continuously think about disrupting the platform and investing in future possibilities, allowing for unplanned changes and adapting to the 'slope' of technological advancement.
30 billion parameters
Model parameter threshold for economic post-training optimization According to a study by Nathan Lambert, once a model hits this size, the CapEx for pre-training with billions of tokens becomes less economically sensible, favoring optimization on the loop.
500
Estimated number of touch points for a product launch In an average organization with 10 steps, 5+ functions, and 6-7 layers, this many touch points are estimated to get a product out.
500 per week
Rate of new AI models/technologies available The rate at which new models or technologies are emerging, making traditional product development insufficient.
70,000
Number of enterprise AI tools launched last year The number of AI tools launched in the enterprise space last year, highlighting the difficulty in choosing the right ones.
15,000
Number of customers building agents on Azure Customers who have produced agents on Microsoft's Azure service.
Millions
Total number of agents running in the cloud on Azure The estimated total number of agents running on Microsoft's Azure service.
3
Average birth rate in the 1990s The average birth rate when Asha Sharma was growing up.
2.3
Current average birth rate The current average birth rate, which is declining.
2050
Estimated year for birth rate to fall below replacement levels The year by which the birth rate is estimated to fall below replacement levels.
50%
Percentage of developers fine-tuning models According to surveys, this percentage of developers are now fine-tuning models.