Bolt.new tutorial builds a three-agent research-to-social-post pipeline
A Bolt.new tutorial walks through building a three-agent AI pipeline that generates research questions, runs deep research, and formats the findings into posts for X or LinkedIn — all without leaving the browser.
Score breakdown
The tutorial demonstrates a concrete multi-agent pattern — chaining question generation, deep research, and content formatting into separate agents — that the source describes as reducing hallucinated facts in AI-generated content.
- 01Three AI agents handle the pipeline: question generation, deep research, and social post formatting.
- 02Agent one uses Anthropic's Claude Opus 4.8 model to generate research questions.
- 03Agent two calls the Google Deep Research API and can take up to 20 minutes to return a report.
The Bolt.new team published a tutorial showing how to construct a three-stage, multi-agent social media content pipeline entirely inside Bolt.new, using browser-based WebContainers with no local setup required. The first agent accepts user inputs — role, industry, company, and complexity level — and sends them to Anthropic's Claude Opus 4.8 model to generate targeted research questions. The tutorial notes that Claude Opus 4.8 is described as a new model, and the prompt instructs Bolt.new to research how to implement it. The second agent takes a user-selected question and dispatches it to the Google Deep Research API using the most recent available model, generating a full research report. Because deep research can take up to 20 minutes, the tutorial includes explicit instructions to prevent timeouts and to inform the user of the wait. The third agent then transforms the completed report into a social post formatted for either X or LinkedIn.
The tutorial also covers a debugging step: on the first attempt, the deep research report returned empty, and the walkthrough shows how to prompt the agent to fix the underlying code issue.
The tutorial also covers a debugging step: on the first attempt, the deep research report returned empty, and the walkthrough shows how to prompt the agent to fix the underlying code issue. The broader lesson the tutorial emphasizes is that splitting one large prompt into three focused agent jobs keeps fabricated facts out of the final draft and makes each agent's work visible and auditable. The video includes timestamped sections covering research question generation, the deep research agent, UI improvements, social post generation, and testing.
Key facts
- 01Three AI agents handle the pipeline: question generation, deep research, and social post formatting.
- 02Agent one uses Anthropic's Claude Opus 4.8 model to generate research questions.
- 03Agent two calls the Google Deep Research API and can take up to 20 minutes to return a report.
- 04Agent three formats the research report into a post for X or LinkedIn.
- 05The tutorial is built entirely in Bolt.new using WebContainers — no local setup or servers required.
- 06On the first test run, the deep research report returned empty, and the tutorial demonstrates fixing the code.
- 07The three-agent split is presented as a way to reduce hallucinated facts compared to a single large prompt.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 13, 2026 · 08:58 UTC. How this works →