Code is becoming a disposable byproduct ('sawdust') of the software development process, with primary value shifting to high-quality prompt engineering.
Humans must retain control over high-level architectural and user interface decisions to ensure software feels 'crafted' and maintains a high standard of quality.
The future of software development involves AI agents that are significantly more powerful and operate in persistent cloud environments, not on local machines.
AI models require careful management and human-curated codebases ('slot-free zones') to avoid negative feedback loops where they learn from and perpetuate bad code.
Software applications will become increasingly 'malleable,' allowing end-users to modify them in a manner similar to how video games are modded.
July 2025
During the initial development phase of Conductor, Holtz reports spending $22,000 on tokens in a single month, indicating a period of intense, high-cost experimentation.
Summer 2024
Conductor is accepted into and participates in the Y Combinator Summer 2024 batch, marking a key milestone in its startup journey.
Present
Holtz describes Conductor as a Mac-native desktop application built with Tauri and Rust, which his team uses to build the product itself. He identifies the current limitation that agents stop when a user's laptop is shut down.
Near Future
Holtz is actively investing in and planning for a significant shift to cloud infrastructure. This is driven by his prediction that agents will soon become 10x smarter and require persistent, powerful environments beyond a local Mac's CPU.
▶The Obsolescence of CodeJun 2026
Holtz posits that the act of writing code is becoming a secondary artifact, or 'sawdust,' in the software development lifecycle. He believes the primary work is shifting to crafting effective prompts that guide AI agents, predicting that entire codebases will become disposable and easily regenerated by improved AI models.
This theme suggests a fundamental shift in the valuation of developer skills, prioritizing high-level prompting and architectural design over line-by-line coding proficiency, which could disrupt developer hiring and training paradigms.
▶Human-Centric AI OrchestrationJun 2026
Despite his bullish stance on AI capabilities, Holtz advocates for a development model where humans retain ultimate control. He argues for human oversight on architectural decisions and UI to ensure a 'crafted' feel, and implements 'slot-free zones' to prevent AI-induced quality degradation from negative feedback loops.
Investors should note that this philosophy positions Conductor not as a full automation tool, but as a human-AI collaboration platform, which may appeal to enterprises wary of ceding complete control to AI.
▶Pragmatic Tooling and Infrastructure Scaling
Holtz details his specific preferences for AI models, using Opus for creative tasks and Codex as a 'workhorse' for debugging. He acknowledges the limitations of local hardware, evidenced by his own 128GB RAM machine and Conductor's current desktop constraints, which drives his strategic investment in a more powerful and persistent cloud infrastructure.
Holtz's focus on moving from constrained local environments to the cloud indicates a significant upcoming capital expenditure and a strategic pivot necessary for his vision of more powerful, long-running AI agents to be realized.
▶Dogfooding as a Product StrategyJun 2026
The development of Conductor is deeply intertwined with its own use; the team builds the application using the application itself. This 'dogfooding' approach informs their product decisions, which are based on their own 'gut feel' and daily usage rather than external analytics, leading to user-inspired features like 'Gary Mode'.
This insular, intuition-driven product strategy can lead to a deeply refined tool for power users like themselves, but it risks creating a product with a steep learning curve or a feature set that doesn't align with the broader market's needs.