LearnPrompt vs Surge AI

LearnPrompt teaches a workflow; Surge AI produces data. That's the whole gap. LearnPrompt's units are task cards: each one specifies input, output, failure modes and acceptance criteria, so you can hand a job to an Agent, read the diff, run the tests, and decide whether the lesson belongs in AGENTS.md or SKILL.md. Its 8 paths cover AI coding basics, Claude Code, Codex, Agent engineering, Agent Skills, Loop Engineering, Obsidian AI, and Hermes/OpenClaw long-running agent architecture. The genuinely differentiating material is the boring-sounding operational stuff: the Harness 五组件排查法 for locating the missing layer when an Agent drifts, a CLI/IDE/desktop/Cloud selection comparison between Claude Code and Codex, and a rule that Showcases must be replayed including the out-of-bounds counterexamples. Every tutorial lists a first-party source and a verification date. Surge AI is a different species: a credentialed human workforce (doctors, lawyers, engineers, writers) doing RLHF preference collection, red teaming, multimodal and reasoning-heavy labeling, and off-the-shelf post-training runs. Its evaluation catalog is the loudest part — GDP.pdf (professional document comprehension; OpenAI's GPT-5.6 flagship scored 30.7%), ComplexConstraints (entangled, context-inferred instruction following), HANDBOOK.md (long-context policy following), Chartography (Kaplan-Meier curves, Bode plots), the Tuesday Work Index composite, DAYJOB suites for Healthcare/Finance, and Riemann-bench. One is a knowledge base you read; the other is a service that returns labeled data and numbers.

LearnPrompt is free and permanently so — no tiers, no gating, no upsell path in the data. It's open source on GitHub, maintained by Carl, with old course versions kept archived. Your real cost is time: 47 tutorials, no video, no certificate at the end, and content that assumes you've already run a terminal and Git once. Surge AI can't be priced from the page: pricing_type is contact, and the stated disqualifier for buyers is showing up to a scoping call without a budget and a scoped pilot. That tells you the engagement model — custom programs sized to your labeling volume, workforce specialization, and benchmark needs, negotiated rather than listed. The two numbers never meet. A hobbyist comparing them on cost learns nothing useful; the honest comparison is zero dollars against an enterprise contract that requires procurement. If a listed price is a hard requirement for you, Surge AI is out of scope by definition, and if you're a solo developer, so is its entire buyer profile. Note the one price-adjacent fact in the news: training a 4B model on 1,000 expert-written ComplexConstraints rubrics moved MultiChallenge +10.1 and AdvancedIF +8.4 — efficiency framing aimed at teams weighing data spend against compute.