Apr 6, 2026, 10:30 AM ETInterview, Fireside Chat

OpenClaw, Claude Code, and the Future of Software | Peter Yang on The a16z Show

Peter Yang operates a single OpenCLAW agent named "Zoe," named after his intended daughter's name, with multiple Telegram channels for distinct use cases.
Usage patterns are heavily voice-centric, with Yang treating the agent as a personal companion rather than a utility, engaging in daily pep talks and deep insights retrieval.
Yang implemented a custom three-layer memory system using tools like Toby's qmd search to overcome the agent's default tendency to forget context in memory.md files.
The interface relies on a "janky" but flexible setup allowing users to prompt agents to execute complex tasks, such as initiating live Twilio phone calls.
Yang maintains high transparency, granting Zoe read/write access to personal email, calendars, and file drives via a dedicated Mac Mini email account.
He anticipates OpenAI or similar entities will productize this architecture into a mainstream ChatGPT experience that feels more human and capable of executing tasks directly.

Yang predicts a decline in traditional SaaS usage as agents perform tasks (e.g., updating Google Docs, pulling analytics) that previously required opening specific applications.
Applications designed for "feeling" (entertainment, connection) like TikTok or X are expected to survive longer than task-oriented apps which are susceptible to agent automation.
Despite agent efficiency, personal phone usage remains high due to addiction to social media platforms like X, which Yang uses for a morning briefing feed generated by the agent.
Current agent setups struggle with context switching between distinct human intents (e.g., "flirting" vs. "getting work done") unless separated by multiple conversation channels.

Yang distinguishes between Coda (vibe-coding, chatty, less precise) and Codex (deep thinking, highly accurate, slower "pause" behavior), using each based on the task's nature.
He notes that Codex lacks the quality-of-life integrations of Cloud Code, such as direct screen clipping or seamless Chrome connectivity.
The "casino effect" of coding agents is highlighted, where variable latency and inconsistent output create a variable-reward loop similar to social media feeds.
A growing trend of AI-native startups is replacing paid SaaS internal tools with custom code generated by "vibe coders" to eliminate licensing fees.
Tools like Figma are predicted to evolve from execution-only IDEs into "thinking tools" where agents facilitate trial-and-error design exploration.
Excel is cited as a historical analog for coding agents, representing a low-barrier programming language where the code is abstracted away for high-leverage problem solving.

Yang advocates for companies remaining small (2-3 human product team members) augmented by a fleet of agents to replace large, emotionally taxing organizational structures.
Agents can handle objective negotiations and communications, removing emotional friction from interactions that typically degrade human workplace satisfaction (NPS).
The traditional corporate planning cycle (annual OKRs) is viewed as obsolete, replaced by a hybrid model of "fast" iteration to reach local maxima and "slow" reflection to find the next market opportunity.
There is a prediction that unemployment may become a catalyst for entrepreneurship, allowing individuals to pursue dreams and build bootstrap businesses in high school or during career gaps.
Unlike the "productivity porn" of managing 20 agents, the optimal workflow involves rapid execution followed by deliberate pauses to recalibrate strategy.

The AI era is shifting consumer business models from indirect monetization (ads) to direct payment, driven by high user willingness to pay for high-value agent outcomes.
New business models are expected to combine API interfaces for agent-to-agent interaction with consumption-based interfaces for human transaction logs.
The "agent stack" is emerging as a foundational layer, with critical components like identity, payments, and marketing still being defined as the CLI/MCP landscape matures.
Historical data shows that 100% automation of job functions is rare; most AI tools currently provide a "last 9%" lift where human oversight remains essential.
Forward-looking statements suggest the economy will not simply shrink jobs but shift toward higher productivity, potentially leading to shorter workweeks or expanded human ambition.
Yang posits that while AI may displace roles, human desire and innovation have no ceiling, driving demand for new services and luxuries that sustain job markets.