adhd vs Temporal AI

ADHD operates entirely inside the generative layer. It fans out N parallel branches per run under 15 cognitive frames (the paper names regulator, speedrunner, biology, and $0 budget), forbids cross-branch context sharing during divergence, and explicitly forbids the generator system prompt from evaluating. A single critic pass then scores, clusters, and deepens top-K survivors. The mechanism depends on mechanical generator-critic separation via distinct LLM calls with opposing system prompts — vantage-point reframing, not next-step variation. It is distributed as a skill and an npm package with an eval suite, runs on the Claude Agent SDK and the Codex Agent SDK, and is v0.1 on a 2026-05-25 preprint.

Temporal solves a different layer entirely. Durable execution captures Workflow state at every step with no checkpointing or recovery code. Activities retry automatically with backoff, four timeout classes, and heartbeating. Signals, Queries, and Updates read and mutate running Workflows mid-flight, and Workflow Streams add real-time interactivity with running executions. Durable AI agents run via the OpenAI Agents SDK and Google ADK, with LLM and tool calls as Activities, hosted on Amazon Bedrock AgentCore serverless workers, AWS Lambda (public preview), and GCP Cloud Run (pre-release). SDKs cover Go, Java, Python, TypeScript, .NET, PHP, Ruby, and Rust, whose 1.0 shipped GA. Durable Timers sleep for months, and Saga compensation reads like try/catch.

The overlap is thin: ADHD improves the quality of what an agent considers; Temporal ensures the agent finishes what it started. Neither substitutes for the other.

ADHD is free. There is no paid tier, no seat count, no hosted plan — you get the GitHub reference implementation, the npm package, the skill, and the eval suite, and your only real cost is the extra LLM calls that parallel branching and the critic pass consume. That cost scales with N branches per run, which is why the paper's own framing warns off latency- and cost-sensitive interactive agents.

Temporal uses a freemium model. The open-source server is self-hostable, so infrastructure-only teams can run it without a vendor invoice, paying instead in operational effort. Temporal Cloud is the managed path and is where the compliance and support features live: SSO, audit logging, SOC 2/HIPAA posture, and an SLA-backed support line. The vendor site names OpenAI, NVIDIA, Salesforce, Twilio, Cursor, Replit, Retool, and Lovable as production users — useful ammunition if you are the person arguing internally for a new orchestration layer.

Comparing the two numbers directly is meaningless. ADHD's price is zero dollars plus inference tokens; Temporal's price is engineering time on the OSS path or a Cloud contract whose size depends on executions and support tier. The real budget question for ADHD is 'how many branches can I afford per run,' and for Temporal it is 'how many workflows must survive a crash.'