Skip to content
Lenny's 🏦 Vault

Planning in Seasons for AI Development

by Asha Sharma on August 28, 2025

The shift from static product development to AI-driven product evolution requires new planning approaches and organizational structures to succeed in an era of rapid technological change.

Planning in Seasons Rather Than Fixed Timeframes

  • Microsoft AI's planning approach has evolved beyond traditional roadmapping:

    • Identify the current "season" - defined by secular changes happening in the industry or customer needs
    • Seasons can last 3 months, 6 months, or a year depending on market dynamics
    • Current season: "The Rise of Agents"
    • Previous seasons included prototyping AI and reasoning models

  • Three-layered planning structure:

    • Align everyone on the ethos of the current season
      • What are the secular changes?
      • What customer problems need solving?
      • What does winning look like?
      • What is the north star metric?
    • Set loose quarterly OKRs that put you on path to the season's goals
    • Teams operate in squads with 4-6 week goals for specific problem areas

  • Leave slack in the system for:

    • Unplanned opportunities
    • Investing in "the slope" (long-term trajectory) rather than just the current snapshot
    • Continuously thinking about how to disrupt your own platform

From Product as Artifact to Product as Organism

  • Traditional product development: static artifacts shipped with incremental improvements

    • Come up with insight → solve problem → ship → make small improvements → monitor dashboard

  • AI-driven product development: living organisms that evolve through interaction

    • The key metric becomes the "metabolism" of a product team
    • Ability to ingest data → digest rewards model → create outcomes
    • Products get better with more interactions
    • This becomes the new IP of companies

  • Post-training is becoming more important than pre-training:

    • Once models reach ~30B parameters, it's more economical to optimize on the loop
    • Companies will spend more on post-training than pre-training
    • Fine-tuning existing models for specific outcomes (price, performance, quality)
    • Using your own data, synthetic data, or purchased data

The Shift to Code-Native Interfaces and Agentic Society

  • GUIs are being replaced by code-native interfaces:

    • Text streams connect better with LLMs
    • Focus on composability rather than canvas
    • Product makers need to rewire their mindset

  • The coming "agentic society":

    • Marginal cost of good output approaching zero
    • Exponential demand for productivity and outputs
    • Scaling through agents (embedded in tools and software)
    • Org charts becoming "work charts" - task and throughput-focused
    • Fewer organizational layers needed

Characteristics of Successful AI-Driven Organizations

  • Three patterns of successful AI adoption:

    1. Everyone becomes AI-fluent in daily workflows
    2. Apply AI to existing processes and measure impact
    3. Use AI to inflect growth (improve customer experience, co-create new concepts)

  • Common failures:

    • Doing AI for AI's sake
    • Too many simultaneous projects without a blueprint
    • Not treating AI as a real investment (lacking measurement, observability, evaluations)
    • Not building on a platform that allows swapping technologies

  • The rise of the "polymath" and full-stack builders:

    • Traditional organizations: 10 steps × 6-7 functions × 6-7 layers = 500 touchpoints to launch
    • This is insufficient in an era with 500 new models weekly
    • Full-stack builders provide velocity and throughput
    • Focus on the loop, not the lane
View full episode