AI Projects - July 3, 2026

Week of: July 3, 2026

---

Overview

This week's AI Projects beat is dominated by the AI Engineer World's Fair (AIEWF) takeaways, including debates on agent loops and the tension between automation and human agency. Practical guides on multi-agent systems, autonomous CRM tools, and enterprise automation platforms like UiPath also feature prominently, alongside developer experiments with coding agents and prompt optimization.

Stories

1. AI Engineer World's Fair closes with debate on loops and state of AI engineering

Source: Latent Space Link: https://www.latent.space/p/aiewf-daily-dispatch-locomotives

The AI Engineer World's Fair ended with a debate about loops, a report on the state of AI engineering, and closing keynotes focused on what to build next. The event served as a major gathering for the AI engineering community.

For AI builders, the "great loops debate" signals a maturing conversation about agent architectures—specifically how and when to use iterative loops versus one-shot designs. The state-of-AI-engineering report provides a benchmark for practitioners evaluating their own stacks and approaches.

Impact Analysis: Track the loop-vs-one-shot debate as it will influence agent design patterns for the rest of the year.

2. Vercel's Andrew Qu explains why agents are a new kind of software

Source: Latent Space Link: https://www.latent.space/p/vercel-agents-new-software

The Vercel Chief of Software detailed how its agent framework, eve, was created and why skills, sandboxes, and agent-readable websites now matter. This provides a window into how a major platform company is thinking about agent infrastructure.

For teams building agent-based products, the emphasis on "agent-readable websites" and sandboxed execution environments points to emerging best practices. Vercel's approach suggests that treating agents as a fundamentally new software category—not just chatbots with tools—is gaining traction.

Impact Analysis: Consider how your agent projects might adopt skill-based architectures and sandboxed execution similar to Vercel's eve framework.

3. Adobe experiments with "agentic sites" that assemble pages per visitor

Source: Latent Space Link: https://www.latent.space/p/the-website-of-the-future

Adobe is experimenting with "agentic sites" that generate pages around an individual user's intent. At AIEWF, Carlos Sanchez discussed the Web's future, where websites dynamically assemble content for each visitor rather than serving static pages.

This represents a paradigm shift for content creators and web developers: instead of designing fixed pages, builders may need to design intent-driven templates that agents can compose on the fly. For AI project teams, this opens questions about SEO, analytics, and user experience measurement in a world where no two visitors see the same page.

Impact Analysis: Start exploring how your content or product pages could be structured for agentic assembly, as this pattern may become standard.

4. Multi-agent systems explained as the next automation paradigm

Source: Zapier Link: https://zapier.com/blog/multi-agent-systems

Zapier published a complete guide to multi-agent systems, framing them as teams of specialized AI agents that share information and delegate tasks—analogous to high-performing human teams. The guide covers how agents can specialize and collaborate on complex workflows.

For automation builders, this is a practical primer on moving beyond single-agent setups. The guide's accessible framing makes it useful for teams evaluating whether to adopt multi-agent architectures for their own projects, particularly in customer-facing or operational workflows.

Impact Analysis: Use this guide as a starting point for designing multi-agent systems that delegate tasks across specialized agents rather than relying on one monolithic model.

5. Simon Willison releases llm-coding-agent 0.1a0 and optimizes Datasette Agent prompts with DSPy

Source: Simon Willison Link: https://simonwillison.net/2026/Jul/2/llm-coding-agent/#atom-everything

Simon Willison released an early alpha of llm-coding-agent, a coding agent built on his LLM tool. Separately, he documented using DSPy to evaluate and improve Datasette Agent's SQL system prompts, demonstrating a systematic approach to prompt optimization.

These are hands-on, open-source projects that show how individual developers are building and refining AI agents. The DSPy-based prompt evaluation workflow is particularly valuable for teams looking to move beyond manual prompt engineering toward data-driven optimization.

Impact Analysis: Follow Willison's llm-coding-agent releases for a lightweight, open-source coding agent reference implementation.

6. Skill engineering and the case against one-shot AI design

Source: Latent Space Link: https://www.latent.space/p/skill-engineering-design

Paul Bakaus discussed Impeccable, human judgment in a "loopmaxxing" era, and why agents still need people to steer them. The piece argues against one-shot AI design in favor of skill engineering—building reusable, human-guided capabilities.

For project teams, this reinforces that the most effective AI systems are not fully autonomous but rather designed with human-in-the-loop feedback. The "skill engineering" concept offers a framework for building agent capabilities that can be composed and governed by human operators.

Impact Analysis: Evaluate your projects for opportunities to implement skill engineering patterns that keep humans in the steering loop.

7. Autoresearch and the tension between AI and human agency at AIEWF

Source: Latent Space Link: https://www.latent.space/p/aiewf-daily-dispatch-agency

The "software factory" vision at AIEWF met resistance from speakers defending human understanding and control. The tension between full automation (autoresearch) and preserving human agency was a central theme of the conference's daily dispatch.

This debate has direct implications for AI project design: teams must decide where to draw the line between automated and human-driven processes. The pushback against the software factory vision suggests that users and builders alike value transparency and control, even as automation capabilities advance.

Impact Analysis: When designing AI projects, explicitly define which decisions remain human-controlled and which are delegated to agents.

Source Links

• Latent Space - AIEWF Daily Dispatch: The great loops debate and the state of AI engineering
• Latent Space - Vercel's Andrew Qu on why agents are a new kind of software
• Latent Space - The website of the future may assemble itself for every visitor
• Zapier - What is a multi-agent system? A complete guide
• Simon Willison - llm-coding-agent 0.1a0
• Latent Space - Skill engineering and the case against one-shot AI design
• Latent Space - AIEWF Daily Dispatch: Autoresearch and the tension between AI and human agency