Divya van Mahajan

The Biggest Trends from the AI Engineer World's Fair

ai-daily-brief-podcast

2025-06-07: The Biggest Trends from the AI Engineer World’s Fair

Overview

This episode of the AI Daily Brief (hosted by NLW) covers the major themes and trends emerging from the AI Engineer World’s Fair in San Francisco (June 2025), a three-day conference organised by Swix (Shawn Wang) and the Latent Space / AI Engineering Summit team. The central argument is that the AI Engineer World’s Fair provides one of the clearest previews available of where agents, tooling, and AI-native software development are headed — making it relevant not just for engineers but for anyone trying to understand the near-term future of AI. The episode also includes a short headlines segment covering fast-growing AI software vendors, ElevenLabs v3, Cursor’s growth, and Higgsfield.

Source video: (URL not provided; search for “AI Daily Brief 2025-06-07 biggest trends AI Engineer World’s Fair” on YouTube)

Prerequisites

  • Familiarity with the concept of AI agents and agentic workflows (autonomous or semi-autonomous AI systems that take sequences of actions)
  • Basic understanding of LLMs (Large Language Models) and products like ChatGPT, Claude (Anthropic), and Gemini (Google)
  • Awareness of MCP (Model Context Protocol) — Anthropic’s open protocol for connecting agents to external tools and data sources
  • General knowledge of the AI developer/builder ecosystem (coding assistants, workflow automation tools, etc.)
  • Familiarity with terms like ARR (Annual Recurring Revenue), evals (evaluation frameworks for AI outputs), and RAG/retrieval systems is helpful

Main Points

Headlines: Agent Companies Dominate Fast-Growing Software Vendors

  • Ramp (corporate card/expense platform) publishes monthly data on fastest-growing software vendors based on business spending.
  • June 2025 top five: Google One, Anthropic, Descript, N8N, and Lindy — the majority are agent or agentic workflow companies.
  • N8N is valued for customisability, including human-in-the-loop review steps; Lindy is used for personalised sales outreach at scale.
  • Interpretation: agents are no longer a future concept — real businesses are paying for them now.

Headlines: ElevenLabs v3 Alpha, Cursor’s Growth, and Higgsfield

  • ElevenLabs 11v3 Alpha launches as their “most expressive” TTS model: 70+ languages, multi-speaker dialogue, and a new Audio Tag system (e.g., [excited], [whispers], [laughs]) for fine-grained emotional control.
  • Cursor crosses $500M ARR (2.5× growth since March), valued at ~$9.9B in its latest round.
  • Higgsfield (AI video generation) grew from $0 to $11M ARR in 8 weeks by offering practical controls — camera angles, consistent characters, cinematic shots — oriented toward immediate use cases like ad production. Represents the shift toward application-layer AI tooling.

The AI Engineer World’s Fair: Scale and Structure

  • A dense three-day event in San Francisco with 20+ tracks running simultaneously, covering: agent reliability, MCP, evals, infrastructure, retrieval/search, security, voice, vibe coding, software engineering agents, robotics, generative media, tiny teams, AI product management, Fortune 500 AI, and more.
  • Host and co-organiser Swix (Shawn Wang of Latent Space) gave a mini-keynote titled “Designing AI-Intensive Applications.”
  • Content is freely available on YouTube at youtube.com/@ai.engineer.

Theme 1: Agents (Broadly Defined)

  • Agents were the dominant theme across multiple tracks: agent reliability, software engineering agents, MCP, and voice.
  • Voice was called out by Swix as “the hottest modality” — finally good enough for production use, with OpenAI presenting on building voice agents.
  • Swix’s key framing slide: “The value of the AI product is in the value of the AI leverage on your effort. Doesn’t matter how agentic, just increase the ratio of human input to valuable AI output.”
  • LLM call ratios are shifting: from 1:1 (ChatGPT-style) → 1:100 (deep research, Codex) → 0:N (ambient/proactive agents).

Theme 2: Infrastructure and Building

  • Key tracks: MCP, infrastructure, retrieval/search, security, evals.
  • Anthropic issued a Request for Startups (RFS) at the MCP track, calling for:
    • MCP servers beyond dev tools (sales, finance, legal, education)
    • Simplified server-building tooling (enterprise and indie-grade)
    • Automated MCP server generation
    • Improved AI security, observability, and auditing
  • Security emerged as a newly serious topic — framed as more practically important to Fortune 500 adoption than many higher-profile tracks. Topics included code-executing agent safety, open standards for agent security, and CISO-approved agent fleet architecture.
  • Attendee energy was heavily oriented toward the workshops and builder sessions, not just keynotes.

Theme 3: New Ways of Working — Tiny Teams

  • A dedicated “Tiny Teams” track featured companies executing large-scale products with very small headcounts.
    • Gumloop: path to becoming a 10-person unicorn.
    • Gamma: small team using agents to multiply output.
  • Swix proposed a metric: companies with more millions in ARR than employees as a working definition of a successful tiny team (implying per-employee revenue > $1M, making profitability accessible without VC dependence).
  • New roles highlighted: AI Architect and AI Product Manager as emerging disciplines.

Theme 4: Coding Agents and Multi-Agent Systems

  • Dedicated tracks for vibe coding and software engineering agents; standing-room-only attendance for the Cursor/Windsurf-style tooling keynotes (Kevin Howe, Windsurf head of product engineering).
  • OpenAI co-founder Greg Brockman argued that the AGI-era future looks like a “menagerie” of specialised, domain-specific agents working together — not a single monolithic AI.
    • Key quote: “The evidence has really been shifting towards this menagerie of different models… there’s a lot of power to be had by models that are actually able to use other models.”
  • Google PM for AI coding (Google Labs) presented a session titled “Your Coding Agent Just Got Cloned and Your Brain Isn’t Ready” — focused on the shift from sequential, synchronous coding to orchestrating parallel agents.
  • This shift is framed as simultaneously a mindset shift, an organisational design shift, and an operational shift.

Three Areas Where AI Engineers Are “Ahead of the Curve”

  1. Evals — Evaluations are increasingly seen as a core competitive moat. Quotes from the broader ecosystem:

    • Gary Tan: “Evals are emerging as the real moat for AI startups.”
    • Kevin Weil (OpenAI CPO): “Writing evals is going to become a core skill for product managers.”
    • Mike Krieger (Anthropic CPO): “Writing evals is probably the most important thing.”
    • Greg Brockman: “Evals are surprisingly often all you need.”
    • The AI Engineer World’s Fair held a dedicated evals track for the first time.

  2. Tiny Teams as a discipline — Formalising the idea that AI-leveraged small teams can achieve outsized output; profitability becomes achievable without traditional VC scaling.

  3. Multi-agent system architecture — The community is already designing for environments with many coordinated agents, not just optimising single-agent capability.

Key Concepts

  • AI Engineer World’s Fair — Annual flagship conference for the AI engineering/builder community, organised by Swix (Shawn Wang) and the Latent Space team.
  • MCP (Model Context Protocol) — Anthropic’s open protocol enabling agents to connect to and interact with external tools, data sources, and servers.
  • Evals (Evaluations) — Structured frameworks and test suites used to measure AI system output quality, reliability, and task performance; increasingly treated as a strategic product asset.
  • Agentic workflow — An AI-driven process where an LLM or agent takes multiple sequential or parallel actions with limited human intervention to complete a task.
  • Multi-agent system — An architecture in which multiple specialised AI agents collaborate, delegate, or operate in parallel to accomplish complex tasks.
  • Tiny Team — A small company (or team within a company) that achieves high revenue or output by leveraging AI agents to multiply individual productivity; Swix’s proposed metric: ARR in millions ≥ number of employees.
  • Vibe coding — Informal term for AI-assisted, often natural-language-driven software development where the programmer iterates rapidly with AI-generated code.
  • Audio Tags (ElevenLabs v3) — Metadata tokens inserted into TTS prompts (e.g., [laughs], [whispers]) to direct expressive vocal output.
  • N8N — Open-source workflow automation platform supporting agentic workflows with customisable human-in-the-loop steps.
  • Lindy — AI-powered automation tool used for personalising sales and business workflows.
  • Higgsfield — AI video generation startup offering production-ready controls (camera angles, character consistency, cinematic shots) for advertising and creative use cases.
  • RFS (Request for Startups) — A public call from an investor or platform (here, Anthropic) outlining specific product areas they want to see built.

Summary

The AI Engineer World’s Fair 2025 presented a detailed and coherent picture of where the AI builder community is focused: the conference confirmed that agents — broadly construed as any system that dramatically increases the ratio of AI output to human input — are already real and commercially deployed, not merely a future aspiration. Four overlapping themes dominated: the maturation of agents (especially voice), the infrastructure needed to support them (MCP, security, evals), new organisational models made possible by AI (tiny teams, new PM and architect roles), and a fundamental shift in how software is built (from single-agent coding assistants to orchestrated multi-agent systems). The host argues that three areas in particular represent leading indicators worth tracking: evals (rapidly becoming the core professional skill and competitive moat in AI product development), tiny-team operational models (a new discipline in high-leverage small-team execution), and multi-agent system design (the architectural paradigm the engineering community is already building toward, even as mainstream discourse still focuses on individual agent capabilities).