Interview, Fireside Chat
From Software Engineers to AI Word Artisans: Filip Kozera of Wordware
- Philip Kizera, co-founder of WordWare, positions the company's mission as bridging human creativity and AI by treating English as the "assembly language" for Large Language Models (LLMs), rather than the primary programming language itself.
- The "Word Artisan" or "WordWare Engineer" is defined as a role focused on structuring English with precise logic to encode human taste and intent into AI systems.
- WordWare distinguishes itself from "no-code" tools by maintaining a structured editor that incorporates programming concepts such as loops, conditional statements, and function calling.
- The platform provides three distinct deployment methods:
- As an API to power complex AI features within existing products.
- As an AI-native workflow engine for complex reasoning tasks beyond simple prompt chaining.
- As a "GitHub for AI," enabling users to share, fork, and build upon pre-existing agent components.
- Current user adoption focuses on "analytical creatives" (e.g., CEOs, technical PMs) who can define clear intent and structure for AI agents, rather than fully non-technical users.
- WordWare's user interface has evolved from a 2D canvas block system to a "programmable document" format to manage the complexity of nested reflection loops and complex agent logic.
- The company argues that "taste" and specific human intent will remain the defining differentiators in a post-AI world, as machines cannot replicate the neural response to human creative intent.
- Kizera cites the example of Excel's impact in the 1980s to predict that WordWare aims to democratize AI for the next 500 million to 1 billion users, moving them from using raw chat interfaces to encoding repeatable workflows.
- Current AI deployment constraints require users to provide trusted, curated data sources rather than allowing unrestricted internet searches, mirroring the limitations of managing an "intern."
- In a lightning round, Kizera expressed the contrarian view that pre-training data volume and specific model origins (e.g., DeepSeek) matter less than the effective application of the best available models.
- Kizera identified Gemini 2.0 Pro as a standout frontier model, specifically highlighting its ability to ingest massive PDF documents (up to 6,000 pages).
- The speaker disagrees with the notion that pre-training is hitting a wall, arguing that model intelligence scales logarithmically with resources but yields exponential utility gains.
- Kizera predicts the next major interface paradigm will move beyond chat interfaces toward "programmable documents" that allow users to zoom from high-level intent down to specific logical components.
- Personal context-aware applications are identified as a key mainstream category for the near future, with AI acting on the context of a user's entire life to make decisions.
- Future AI interactions are expected to transcend simple command-line or chat inputs, utilizing a document-based medium where users can define, iterate, and refine agent behavior through structured narratives.
- The ultimate vision involves a shift where humans act as "CEOs" setting strategy and taste, while AI executes the operational details of knowledge work.