Garrett Lord

CEO of Handshake

Claims

As speaker

Mentioned in

Key positions and views

The primary bottleneck for advancing frontier AI has shifted from pre-training on massive, general datasets to post-training with specialized, high-quality data from human experts.

In the human-generated AI data market, the only defensible long-term competitive advantage is proprietary access to a large, engaged community of experts, not technology or process.

AI models have surpassed the capabilities of human generalists, making subject matter experts in fields like law, accounting, and STEM essential for the next phase of AI improvement.

The current market for high-quality AI training data is supply-constrained, with effectively unlimited demand, allowing providers to sell nearly all the quality data they can produce.

A direct-to-lab business model is superior to working through intermediaries, as it allows for better collaboration and captures more value from the expert network.

Podcast consensus on Lord

Points of consensus

▶Multiple claims and quotes assert that frontier AI models have exhausted the utility of generalist data and now require post-training data from domain experts to improve reasoning, a market shift Handshake is positioned to serve.Apr 2026

▶The scale and composition of Handshake's network, cited consistently as 18 million users including 500,000 PhDs and 3 million master's students, is presented as the company's primary competitive advantage or 'moat' in the AI data market.Apr 2026

▶The demand for high-quality, expert-generated AI training data is characterized as 'unlimited,' enabling Handshake's new AI division to experience explosive growth, evidenced by turning down a $20 million project due to capacity constraints.Apr 2026

▶Experts providing data on Handshake's platform are highly compensated, with multiple claims citing average rates of over $100 per hour, and specific rates of $125/hour or up to $200/hour for PhDs.Apr 2026

Points of debate

▶There is a minor inconsistency regarding the number of frontier AI labs Handshake works with; Lord states they work with 'six' labs in two claims and one quote, but 'seven' labs in another claim and quote.Apr 2026

▶The description of Handshake's user base varies slightly across statements, being referred to as '18 million people,' '18 million professionals,' and '18 million young people,' which could suggest different framing for different audiences.Apr 2026

▶While expert compensation is consistently reported as high, the exact figures vary slightly across claims, described as 'between $100 and $200 per hour,' an 'average of over $100 per hour, with a typical rate of $125 per hour,' and an 'average bill rate of over $100 per hour.'Apr 2026

Key themes

▶The Strategic Pivot to an AI Data BusinessApr 2026

Lord details Handshake's rapid and decisive entry into the AI data market. This involved incubating a new, independent business unit, dedicating over 80% of his own time as CEO, and strategically cutting off intermediary data companies to work directly with AI labs.

This theme demonstrates that established companies with unique assets (in this case, an expert network) can create significant enterprise value by repurposing them, but it requires intense leadership focus and a willingness to disrupt existing business relationships to capture a new, high-growth opportunity.

▶The 'Post-Training' Inflection Point in AIApr 2026

Lord consistently argues that AI development has hit a wall with pre-training on general internet data. He claims the majority of performance improvements now come from post-training processes, which require nuanced data from human experts to enhance model reasoning in specialized fields.

This positions the value chain as shifting from raw data scale to data quality and specificity, suggesting that companies with access to curated, high-expertise communities will hold significant leverage over AI labs in the coming years.

▶Market Dynamics of a Supply-Constrained ResourceApr 2026

Lord portrays the market for high-quality, expert-generated AI data as having 'unlimited demand.' He asserts that the only sustainable moat is not technology but proprietary access to an audience, contrasting Handshake's organic network with competitors' high-cost expert acquisition strategies.

This framing suggests the AI data sector is currently a resource-driven market akin to commodities, where the primary challenge is scaling supply, not finding buyers, making exclusive access to expert labor the most critical factor for success.

▶The Future of Human-in-the-Loop AIApr 2026

Lord predicts that human experts will be indispensable for AI training for at least the next decade, up until the point of Artificial Superintelligence (ASI). While acknowledging the utility of synthetic data in verifiable domains, he maintains it cannot replace human-generated data for complex reasoning tasks.

This long-term view provides a clear investment thesis for Handshake's AI business, framing it not as a temporary bridge technology but as a foundational, decade-long component of the AI development lifecycle.

Source episodes

Sentiment over time

Not enough data for timeline

Changes over time

18-24 months ago

Lord claims that performance gains from pre-training AI models on internet data began to asymptote, creating the market need for high-quality, expert-led post-training data.

Pre-Launch (Recent Past)

Handshake made the strategic decision to stop providing its expert user data to intermediary data companies and pivot to a direct-to-lab model.

January (Current Year)

Handshake officially launched its new AI data business, with Lord dedicating over 80% of his time to the initiative.

Four months post-launch

The new AI business reportedly achieved a $50 million Annual Recurring Revenue (ARR) run rate.

Present

The AI division is on pace to exceed $100 million ARR in its first year, is working with 6-7 frontier labs, hiring aggressively, and turning down multi-million dollar projects due to being at capacity.

Suggested prompts

How sustainable is Handshake's 'moat' of audience access if competitors with larger capital reserves decide to aggressively subsidize expert acquisition on other platforms? &nearr;What are the operational and quality control challenges of scaling from hiring 12 people a week to 20, and how might this impact the quality of the data provided to AI labs? &nearr;Given the claim that models are 'better than generalists,' what is the long-term career outlook for the millions of non-expert students and professionals in Handshake's core network? &nearr;How does Handshake's direct-to-lab strategy for AI data affect its relationship with the 1 million companies, including the Fortune 500, that use its platform for traditional recruiting? &nearr;

Key concepts

AI Post-Training 3 ep Expert Networks 3 ep Handshake Business Model 3 ep AI Data Market 3 ep Frontier AI Labs 3 ep Corporate Recruiting 2 ep University Partnerships 1 ep GPU Infrastructure 1 ep Synthetic Data 1 ep Small Business (SMB) Market 1 ep

Notable quotes

“We like to say the only moat in human data is access to an audience.”

Garrett Lord · Inside the expert network training every frontier AI model | Garrett Lord

“in this market, there's, there's essentially like unlimited demand. Like if you can produce high quality volumes of data, uh, you most likely will be able to sell whatever you produce.”

Garrett Lord · Inside the expert network training every frontier AI model | Garrett Lord

“we just turned down a 20 million dollar project like yesterday morning because we just can't we can't deliver on it”

Garrett Lord · The Expert Network Behind Handshake AI’s Model Training w/ Garrett Lord & Mamoon Hamid

“But really what's happened is the models have gotten so good that the generalists are no longer needed. Like what they really need is experts, experts across every area that the models are focused on.”

Garrett Lord · Inside the expert network training every frontier AI model | Garrett Lord

Report last updated: Apr 21, 2026

Get started free

Back to Entities Intelligence Report