Handshake Turns Expert Networks Into AI Training
💡Handshake's $1B AI pivot reveals what Agent training data looks like after prompts and answers.
⚡ 30-Second TL;DR
What Changed
Handshake's AI business reportedly reached about $1 billion in revenue and has a projected path toward $2 billion.
Why It Matters
This signals a shift from simple data labeling toward full-stack post-training infrastructure for agents. Companies building domain-specific AI may need to invest in expert recruitment, environment simulation, trajectory data, and reliable reward signals rather than relying only on prompt-response datasets.
What To Do Next
Use the open-source Gandalf evaluator to test one production-like Agent workflow with real files and tool calls before expanding your post-training dataset.
Key Points
- •Handshake's AI business reportedly reached about $1 billion in revenue and has a projected path toward $2 billion.
- •The company uses its professional network of more than 30 million people and about one million companies to source verified experts.
- •AI training increasingly requires simulated work environments containing files, software tools, workflows, and success criteria.
- •Handshake acquired Cleanlab to strengthen RL environments, evaluations, AI safety, and human-data capabilities.
- •Its open-source Gandalf evaluator runs in the same environment as an Agent and checks files, databases, and tool outcomes.
🧠 Deep Insight
Background and context from public sources — not the original article. 8 sources cited.
🔑 Enhanced Key Takeaways
- •Handshake operates an 'AI Fellowship' program that compensates PhDs and domain specialists up to $100/hr to generate complex reasoning chains and preference rankings.
- •The company's pivot leverages its historical academic roots to specialize in high-barrier scientific domains, specifically biology, chemistry, and drug discovery.
- •Handshake utilizes a remote, asynchronous operational model to scale its expert workforce dynamically in response to the fluctuating data requirements of frontier AI labs.
- •The acquisition of Cleanlab serves as a critical infrastructure layer, enabling automated label auditing and data quality assurance for the human-generated datasets.
- •Handshake's competitive moat is built on a decade-long trust relationship with university students and alumni, providing a unique pipeline of graduate-level talent in physics, mathematics, and computer science.
📊 Competitor Analysis▸ Show
| Feature | Handshake | Scale AI | Labelbox |
|---|---|---|---|
| Primary Source | Academic/Expert Network | Crowdsourced/Managed | Enterprise Platform |
| Domain Focus | Scientific/Professional | General/Computer Vision | Data Management/Ops |
| Pricing Model | Expert-based/Project | Volume/Task-based | SaaS/Subscription |
| Key Differentiator | High-trust academic talent | Massive scale/RLHF | Data quality tooling |
🛠️ Technical Deep Dive
- Implementation of Reinforcement Learning (RL) environments that simulate professional workflows to train agentic reasoning.
- Integration of Cleanlab's automated auditing to detect and correct label noise in human-generated training sets.
- Deployment of the Gandalf evaluator, an open-source framework that executes within the agent's sandbox to verify tool outcomes and database interactions.
- Generation of multi-step reasoning chains designed to improve model performance in complex, multi-domain scientific tasks.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (8)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.