What's Easy Now? What's Hard Now?

Key Takeaways

• Coding agents excel with precise, automated feedback loops
• UI/UX projects stay hard due to human‑dependent feedback
• System software like databases becomes easier for agents over time
• Specification tools (Rust, TLA+, Verus) are critical for success
• Future dev tools will focus on building robust feedback mechanisms

Pulse Analysis

The rise of AI‑driven coding agents has sparked excitement, but their true power lies in how they receive and act on feedback. Traditional open‑loop models, such as autocomplete, rely on developers to interpret errors and iterate manually. By embedding the build‑test‑refine cycle inside the agent, developers can offload repetitive debugging and let the model converge on correct implementations faster. This shift mirrors the feedback principle long used in electronics, where a simple component’s performance is amplified through a loop that corrects its output.

When feedback is quantifiable—through type systems, property‑based tests, or performance benchmarks—agents can autonomously improve code quality. Languages like Rust, with strict compile‑time checks, and tools such as Hydro or Verus, provide the deterministic signals needed for rapid convergence. Conversely, tasks that depend on subjective user experience, like designing an ergonomic photo‑editing web app, lack machine‑readable criteria, forcing the loop to involve slow, inconsistent human judgments. As a result, these UI‑heavy projects remain challenging for AI, despite their apparent simplicity.

The long‑term implication is a strategic pivot for software firms. Investing in formal specifications, model‑checking, and automated verification will unlock AI’s full potential, especially for infrastructure‑level components such as high‑performance databases. Companies that prioritize building robust feedback ecosystems can expect faster development cycles, lower costs, and a competitive edge as AI agents take on increasingly complex engineering problems. The feedback loop hypothesis thus redefines what is "easy" for AI and signals a new era of specification‑centric software engineering.