Most Developers Call AI Data With APIs and A2A

Source: The New Stack

AI Development Survey Highlights

A recent Theory Ventures survey found that 91% of 413 senior technical builders surveyed are either directly building AI or managing teams that build AI. Only 17% are building MCP servers, while 46% are using direct tool calling—including via APIs and A2A—to access data.

• Most common use case for agent tools: accessing databases (72% of respondents).
• Quality evaluation: 98% are evaluating AI quality, typically with a mix of automated and manual methods.
  • 27% use LLMs to judge output quality.
  • 63% use synthetic data for evaluation.
• Direct tool calling: 46% use it to access data via APIs and A2A.
• AI evaluations vs. telemetry: Respondents are almost three times more likely to use AI evaluations to assess product quality than telemetry.
• LLM observability: 57% store interactions as traces in an observability product, and another 57% also use spreadsheets for data review.
• Tooling value: 21% believe data review of interactions will add the most value.
• Context engineering:
  • 47% use prompt‑optimization methods like GEPA and Prompt Evolution.
  • 52% manage prompts as code in repositories with code‑review processes.
  • 41% update prompts via collaborative text files.
  • An additional 41% iterate on prompts outside of code.

Shipaton Mobile Hack‑A‑Thon Won With Vibe‑Coded App

Shipaton is a mobile‑app hackathon hosted annually by RevenueCat, a platform that powers in‑app purchases. This year’s event highlighted the rise of “vibe coding.”

• Participation: 1,700 participants last year → 54,000 this year.
• Grand Prize: Payout – an AI‑assisted app built with Claude Code and Cursor that helps users find qualifying class‑action lawsuits.

“AI tools are lowering the barrier to build and ship apps, but the fundamentals haven’t changed. You still need to build something people love, use, and pay for,” — Jacob Eiting, CEO of RevenueCat.

Award Categories & Winners

• Best Vibes: OtterDay – used Perplexity Pro (dialogue & visuals), KlingAI (otter animations), ElevenLabs (voiceovers).
• Design: DayLoop – creates cinematic time‑lapse videos from everyday moments, emphasizing precision, privacy, and delight.
• Buzziest Launch: ReadHim – decodes men’s texts; achieved 5.2 M Instagram views, partnered with a TikTok influencer (2.3 M followers), and executed a stunt with supercars and a robot dog.
• Apps That Make Money: VectorGuard – recognized for a sustainable monetization strategy that turns public data into public good.

RevenueCat also presented six Shippies awards for apps excelling across the subscription journey (onboarding, monetization, retention, creativity). 2025 Shippies recipients: Focus Friend, Ladder, Resubs, Recime, Wink, and WeWard.

Warp CLI Tool Expands AI Agent Capabilities

Warp is a modern, high‑performance command‑line terminal for macOS, Linux, and Windows. This month Warp released Agents 3.0, extending its AI agent capabilities.

New Features

• Full Terminal Use: Agents can interact with live processes and full‑screen terminal apps (e.g., debuggers), removing a major bottleneck for real‑world development tasks.
• Structured, Versioned Development Blueprints: Provide reproducible, version‑controlled AI‑generated code plans.
• Interactive Code Review: Enables human oversight directly within the terminal, allowing developers to approve, modify, or reject AI‑suggested changes on the fly.

These enhancements aim to deliver reliable, collaborative, and fully autonomous development workflows inside the terminal environment.