# Agent Labs vs. Model Labs — Why the Shift Is Happening Now
*[AIE CODE](https://www.ai.engineer/code) is sold out, but you can [watch the livestream](https://youtube.com/live/cMSprbJ95jg?feature=share), join [AIE CODE++ in SF](https://luma.com/aieng), or attend the [Dev Writers Retreat](https://lu.ma/dwr2025) after NeurIPS!*
---
## Background: From “Neolab” to “Agent Lab”
We admire [Steph Palazzolo](https://www.youtube.com/watch?v=t4IQwMa5-6U&t=3945s), but we respectfully disagree with her [“Neolab” term](https://www.theinformation.com/articles/investors-chase-neolabs-outflank-openai-anthropic?rc=ytp67n), amplified by [guest Deedy Das](https://www.youtube.com/watch?v=8UDj3-JDfYY&t=7s).
At the AI Engineer Code Summit, we introduced the term **Agent Lab** — reflecting the evolving role of autonomous agents and creator-driven experimentation, rather than focusing solely on “new” model research.
Platforms like [AiToEarn官网](https://aitoearn.ai/) exemplify this vision, providing **open-source tools** for AI content generation, multi-platform publishing, analytics, and monetization across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X.
[](https://substackcdn.com/image/fetch/$s_!fTkr!,f_auto,q_auto:good,fl_progressive:steep/.../54bac27a-39b9-462f-926e-dc84b45efec6_1960x1740.png)
Visit: [https://ai.engineer/code/2025](https://ai.engineer/code/2025)
---
## What Defines an Agent Lab?
**Agent Labs** differ from Steph’s *Neolabs*, which focus on overlooked AI model research.
Agent Labs build **practical, end-user-facing AI agents** and productivity tools.
### Major Agent Labs
- [Cursor ($29B)](http://latent.space/p/cursor)
- [Perplexity ($20B)](https://techcrunch.com/2025/09/10/perplexity-reportedly-raised-200m-at-20b-valuation/)
- [Cognition ($10B)](https://www.swyx.io/cognition)
- [Sierra ($10B)](https://www.latent.space/p/bret)
- [Lovable ($2B)](https://lovable.dev/blog/200m-series-a-fundraise)
- [Gamma ($2B)](https://techcrunch.com/2025/11/10/ai-powerpoint-killer-gamma-hits-2-1b-valuation-100m-arr-founder-says/)
### Established Companies with Agent Integration
- [Notion ($10B)](https://www.cnbc.com/2025/09/18/notion-launches-ai-agent-as-it-crosses-500-million-in-annual-revenue.html)
- [Vercel ($9B)](https://vercel.com/blog/series-f)
- [Glean ($7B)](https://www.glean.com/blog/glean-series-f-announcement)
- [Replit ($3B)](https://x.com/pirroh/status/1938608564203049183)
### Agent Labs Inside Model Labs
- [Claude Code ($1B ARR)](https://www.latent.space/p/claude-code)
- [Codex](https://www.latent.space/p/codex)
---
## Why “Agent Lab” ≠ “Model Lab”
**Key distinction:** Model Labs research and sell models; Agent Labs research and sell agents.
[](https://substackcdn.com/image/fetch/$s_!ur9D!,f_auto,q_auto:good,...)
---
## Model Labs vs. Agent Labs: Best Practices
> *Not all Agent Labs will meet these criteria initially, but they will converge over time due to PMF and economic forces.*
### 1. Product First, Model Last
- *Model Lab approach*: Raise big funding for model R&D before building a product (e.g., [Magic.dev](https://magic.dev/blog/100m-token-context-windows)).
- *Agent Lab approach*: Build and refine product **before** customizing models (e.g., [Cursor](https://www.youtube.com/watch?v=deMrq2uzRKA)).
### 2. Outcome-Based Pricing
- Model Labs: Often constrained by low monthly subscription ARPU.
- Agent Labs: Can command higher tier/outcome-based pricing (e.g., [$2,000/month agents](https://x.com/factoryai/status/1989483223630712966?s=46)).
### 3. Approach to Autonomy
- Model Labs: Focused on maximum autonomy for AGI ambitions.
- Agent Labs: Value speed, auditable human oversight, and iterative harness rebuilding.
### 4. Practical Metrics
- Model Labs: Optimize for max capability benchmarks.
- Agent Labs: Balance intelligence & cost on a Pareto frontier.
### 5. Open Sourcing Internal Agents
- Example: [OpenAI’s released agents](https://openai.com/index/openai-inbound-sales-assistant/).
- Example: [Vercel’s agent learnings](https://vercel.com/blog/what-we-learned-building-agents-at-vercel).
---
## Economic & Strategic Context
### Inference Spend
[Epoch](https://x.com/EpochAIResearch/status/1976714284349767990/photo/1) — only **28%** of OpenAI’s compute goes to inference:
[](https://substackcdn.com/image/fetch/$s_!lFwA!...)
### AI Cloud Pivot
OpenAI, Anthropic, Vercel, GitHub, and now Cloudflare are moving toward **AI cloud strategies** rather than only consumer products.
---
## Recap: Why Agent Labs Now?
- **Priorities:** Model Labs = AGI models. Agent Labs = AI agents.
- **Inference Economics:** Model Labs spend <30% budget on inference.
- **Platform Orientation:** Shifting toward developer ecosystems.
- **Model Diversity:** Many frontier models to orchestrate.
- **Agent Complexity:** Agents = systems (model + prompt + memory + tools).
- **Shift in Training Economics:** Moving from large-scale pretraining → domain-specific reinforcement learning.
---
## The Role of Cross‑Platform AI Monetization
Platforms like [AiToEarn官网](https://aitoearn.ai/) are increasingly relevant:
- **Generate** AI content with integrated tools.
- **Publish** simultaneously to Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X.
- **Analyze** with built-in analytics and [AI模型排名](https://rank.aitoearn.ai).
- **Monetize** globally via open source infrastructure.
---
## The End of “One‑Size‑Fits‑All”?
- Omnimodel ambitions (e.g., GPT‑4o) haven’t yet fully materialized in GPT‑5.
- Specialized models emerging → Increased importance for orchestration platforms like AiToEarn to connect multiple AI tools into unified workflows.
---
## Conclusion
**Agent Labs are here to stay.**
They represent a **product-first, outcome-driven, human‑in‑the‑loop** approach to deploying AI at scale — distinct from the fundamental‑research focus of Model Labs.
**Monetization & distribution platforms** like [AiToEarn官网](https://aitoearn.ai/) will become integral in this ecosystem, ensuring Agent Lab outputs can be **published, tracked, and monetized** efficiently across multiple global channels.
---
