A fireside Chat with May Habib, CEO & Co Founder of Writer ft. Rana Yared, GP, Balderton Capital

Market Context & Shift

• The last six months mark a transition to "Era 3.0" of AI, characterized by the rise of useful, revenue-impacting agentic systems rather than just efficiency tools.
• Enterprise clients are shifting focus from simple information extraction to "agentic premiums," such as instantly cross-populating systems to generate specific revenue figures (e.g., "$2 billion of cross-sell opportunities").
• Writer, a full-stack generative AI company, serves enterprise clients by building agentic solutions ranging from LLMs to WYSIWYG application layers.

Go-to-Market Strategy & Implementation

• Writer's strategy is organized vertically by use cases that demonstrate immediate business value rather than abstract technology.
  • Sales: Enterprise sales planning for accounts (e.g., Qualcomm) to define meeting strategies.
  • Retail: Agentic SEO to ensure products are recommended by consumer LLM agents and to restructure retailer sites automatically.
• Implementation timelines are configured to deliver value within 30, 90, and 365-day increments to maintain executive interest between proof-of-concept and ROI realization.
• "Greenfield" opportunities require a rewiring of corporate mindsets and strategy, not just technology integration (e.g., transforming a wholesale blood-testing business into a DTC model with 5% of the original headcount).

Legacy Systems & Integration Horizon

• Short-term: Legacy system providers will likely offer "agentic interfaces" to their own data rather than replacing the systems entirely; most current legacy architectures lack the capability for true agentic control.
• Medium-term: Enterprises will utilize "shift and lift" strategies, translating standard operating procedures (SOPs) into agent-driven processes through a mix of deterministic and non-deterministic workflows.
• Long-term: As transformers absorb internalized memory, complex integrations (e.g., Snowflake, Qualtrics) may be bypassed entirely, leading to a fundamental restructuring of how enterprises use tooling and databases.

Barriers to Adoption (2025)

• The primary barriers to adoption are identified as Security, Scalability, and Reliability.
• Writer addresses these by investing heavily in retrieval techniques to control probabilistic technology and ensuring high precision for complex use cases.
• Traceability & Auditability: Systems are designed to make the agent's decision path traceable, enabling "CYB" (Cover Your Business/Behavior) and regulatory compliance, which acts as a proxy for trust (similar to "CYA" but for business actions).
• Agent Development Lifecycle: Distinct from traditional software development, this lifecycle includes Git-like functionality for managing prompts, tool configurations, and rules, alongside agent-suggested evaluations and alerts.

Technical Architecture & Frugality

• Writer pioneered the use of synthetic data for training transformers in 2020, establishing a culture of frugality and contrarian technical decisions to maintain low Total Cost of Ownership (TCO).
• The company maintains a full-stack approach (Agent Builder, Agent Interaction/Discovery, Agent Security) to control the entire pipeline, ensuring reliability and consistency while avoiding negative gross margins.
• Writer's models are purpose-built for retrieval and agent interaction, allowing them to outperform competitors on cost while delivering high-accuracy, complex outcomes.
• Future model iterations (e.g., "Palmyra") are expected to require less context and engineering to achieve current levels of complexity and productivity.

Industry Outlook & Differentiation

• A "shakeout" is expected in the AI sector; differentiating factors for surviving companies will include full-stack ownership and the ability to offer complexity, accuracy, and reliability that infrastructure providers cannot replicate.
• The Moat: Continuous iteration speed is the primary competitive moat, facilitated by owning the technology stack rather than competing against infrastructure providers.
• Culture: Success relies on a culture comfortable with rapid model iteration where newer versions supersede previous engineering efforts to maintain productivity and security.