WhoIsCooking Launches to Monitor and Tame Thermal Load Caused by Local AI Agents on Apple Silicon

Key Takeaways

• Targeted AI Process Grouping: Unlike native system monitors that display isolated background processes, WhoIsCooking automatically correlates parent AI agents—including Claude Code, Codex, and Cursor—with all sub-processes, compilers, and test runners they spawn.
• Non-Destructive Thermal Throttling: The tool features a one-click "Cool down my Mac" mechanism that shifts heavy computational workloads from Apple Silicon performance cores to efficiency cores without freezing or terminating jobs.
• Quantifiable Thermal Reductions: Real-world benchmarks shared by the creator demonstrate chip temperatures plunging from roughly 83°C under unrestrained execution to 48°C under throttled conditions on an Apple M5 processor.
• Developer-Friendly Distribution Model: Offered at a flat $5 one-time cost, WhoIsCooking is signed, notarized, requires no account registration, and functions on Apple Silicon systems (M1 through M5) running macOS 14 and above.

In-Depth Analysis

Diagnosing AI Agent Thermal Overhead on Apple Silicon

The rapid rise of local and terminal-driven autonomous programming assistants—such as Anthropic's Claude Code, OpenAI's Codex CLI integrations, and Cursor's agentic workflows—has fundamentally transformed how software engineers interact with personal workstations. However, giving autonomous software agents free rein over terminal commands, language server protocols, package installations, and automated testing suites creates unprecedented thermal friction. When an autonomous agent begins iterating on code, it frequently spins up multiple competing compilers, Node.js scripts, Docker containers, and test watchers simultaneously.

Standard operating system utilities, notably macOS Activity Monitor, were designed for a computing paradigm where users ran discrete desktop applications. In a modern AI agent workflow, Activity Monitor populates dozens of detached, low-level system processes, obscuring which top-level agent initiated which task and offering zero contextual temperature telemetry. As workstation chassis heat up and thermal throttling degrades overall responsiveness, developers are left guessing which specific agent session is overloading the processor. WhoIsCooking addresses this observability gap by surfacing chip temperatures directly in the macOS menu bar or notch area while mapping complex process trees back to their originating AI agent session.

Process Throttling Without Interruption: The Efficiency Core Strategy

The centerpiece of WhoIsCooking is its non-destructive approach to process management. In high-stakes development scenarios, abruptly killing rogue processes can corrupt local databases, invalidate build caches, or abruptly terminate multi-step code generation tasks. The conventional response—either letting the machine thermal-throttle at peak fan speeds or force-quitting the application—presents a false dilemma that wastes developer time and hardware performance.

WhoIsCooking circumvents this trade-off through architectural thread re-assignment. By executing a targeted core re-prioritization, the software migrates compute-intensive workloads away from high-power performance cores directly onto Apple Silicon efficiency cores. Instead of freezing execution threads or revoking process licenses, the utility forces processes into an energy-efficient execution tier. In developer demonstrations utilizing an Apple Silicon M5 platform, running four intensive worker processes shifted chip temperatures from an elevated 83°C down to a manageable 48°C. Although execution times scale downward proportionally, ongoing agent runs remain fully intact, allowing background refactoring and test passes to finalize without interruption.

Platform Constraints and Software Transparency

To ensure realistic user expectations, developer G Andrew Wilson has delineated clear technical boundaries for WhoIsCooking. Because the tool relies specifically on Apple Silicon core topology and Unix process affinity capabilities available in recent macOS kernel architectures, it strictly requires Apple Silicon hardware ranging from M1 through M5 chipsets running macOS 14 Sonoma or newer.

Furthermore, the utility explicitly does not modify native Apple system daemons or manage dedicated GPU tasks, maintaining a focused scope on CPU-bound developer tooling and CLI agents. Distributed as a standalone, signed, and notarized binary for a single $5 purchase, WhoIsCooking deliberately avoids telemetry bloat, software subscriptions, and mandatory user accounts, aligning with developer preferences for lightweight, privacy-respecting system utilities.

Industry Impact

The arrival of WhoIsCooking marks an early yet revealing inflection point in how AI agents interact with local computing hardware. As agentic AI architectures shift from remote cloud sandboxes to local client execution—driven by latency demands, proprietary source-code privacy, and cost considerations—workstations will increasingly host semi-autonomous agents that consume system resources unpredictably.

This shift implies that traditional operating system scheduling and performance telemetry tools are no longer fully adequate for tracking AI-assisted engineering workflows. When software agents act as surrogate developers, managing process lifecycles, thermal ceilings, and energy footprints becomes an essential discipline within developer tooling. The enthusiastic reception of WhoIsCooking illustrates that utility software in the era of autonomous programming must bridge the gap between AI autonomy and hardware sustainability.

Frequently Asked Questions

What makes WhoIsCooking different from the default macOS Activity Monitor?

While macOS Activity Monitor displays disconnected system processes such as isolated Node runtimes and compilation threads without tracking their root parent agent, WhoIsCooking automatically correlates all dependent sub-processes into a unified row representing each AI agent session. It also integrates real-time hardware temperature metrics in the notch or menu bar, which Activity Monitor omits.

How does the "Cool down my Mac" feature operate without breaking running code?

Rather than sending termination signals (like SIGKILL or SIGTERM) or completely pausing tasks (like SIGSTOP), the cooling function re-schedules active workloads onto Apple Silicon efficiency cores. This drops chip operating temperatures while permitting tests, installations, and builds to continue executing to completion at a reduced clock rate.

Which hardware platforms and macOS versions are supported?

WhoIsCooking requires an Apple Silicon Mac equipped with an M1, M2, M3, M4, or M5 processor running macOS 14 (Sonoma) or newer. It is restricted to CPU-based tasks and does not throttle dedicated GPU execution or core Apple operating system daemons.