Allen Pike

Allen Pike is co-founder and chief executive of Forestwalk Labs, the Vancouver company developing Powerplant, a collaborative software factory built around coding agents and visible, verifiable results. A former Apple engineer and longtime software founder, he designs AI products around three practical constraints: human attention, interface quality, and the cost of getting useful work done.

From Apple to AI-native software

Pike worked on Apple’s iWork team from 2008 to 2010, then co-founded Steamclock Software, leading the mobile-product studio from its original two founders to a 16-person team. He also co-founded Royal Pixel Service and led development of Two Spies, an Apple-featured mobile strategy game that surpassed one million downloads.

In 2024, he transitioned Steamclock’s operating leadership to a managing director and remained on its board. That year, he founded Forestwalk with Jenn Cooper. Its projects have included ScoutEvals, for evaluating language-model applications; Timberline, an agentic Mac application; and Cedarloop, a meeting assistant that acts on conversations. Powerplant applies those experiments to coordinating coding agents without surrendering product judgment.

• Voice-in, visuals-out interfaces: Spoken instructions carry more context than typing, while visual responses avoid interrupting conversations. Pike’s real-time meeting assistant can recognize discussion of a Slack integration bug, create a Linear issue, and acknowledge the action onscreen.

• Latency as a product constraint: Fluid voice-to-voice interaction demands approximately 200-millisecond responses; visual feedback can remain effective within roughly one second. Pike favors fast models, inference while people are still speaking, background handoffs for heavier tasks, and stable prefix caching to reduce repeated processing.

• Agent evaluation that reflects real use: Probabilistic software encounters inputs that deterministic tests cannot anticipate. Pike’s approach to automated evaluation measures whether model-powered features remain useful as prompts, models, and behavior change.

• Coding agents must show their work: Screenshots and passing tests cannot establish whether an interface feels coherent. Pike advocates preview environments and recorded demonstrations; after Forestwalk spent approximately $10,000 on coding agents in one week, he also outlined practical controls on agent costs, including smaller contexts, model selection, and earlier verification.

Pike also hosts It Shipped That Way, a podcast about building products and teams, and organizes Infer Vancouver, an AI-engineering speaker series.