Bolt.new tutorial covers essential prompting techniques for AI builders
A Bolt.new tutorial walks through key prompting skills — including plan mode, context management, model selection, and tools like screenshot inspiration — to help builders ship faster and waste fewer tokens.
Score breakdown
Developers using Bolt.new can apply these prompting habits — especially plan mode and incremental prompting — to reduce wasted tokens and get outputs that more closely match their intended design on the first pass.
- 01The tutorial uses building a two-sided local marketplace for yoga classes as its working example.
- 02The enhance prompt feature rewrites and restructures a user's input for better agent comprehension.
- 03Plan mode causes the agent to generate a detailed implementation plan and ask clarifying questions before building.
The tutorial, published by Bolt.new on the StackBlitz YouTube channel, uses a concrete example — building a two-sided local marketplace for hosting yoga classes in public spaces — to demonstrate how prompting strategy directly affects output quality and token efficiency. The presenter emphasizes prompting the agent incrementally rather than all at once, arguing that overloading the agent reduces its efficacy. The first skill introduced is the enhance prompt feature, which rewrites and restructures a user's input so the agent can better comprehend it. The second, described as "very powerful," is plan mode, which causes the agent to produce a detailed implementation plan and surface clarifying questions — including edge cases the user may not have considered — before any code is written. The presenter also advises selecting a performance-tier model such as Opus 4.7 when building from scratch, and switching to smaller models only for minor tweaks to conserve tokens.
It also demonstrates using image generation and screenshots as design inspiration, and using version history to roll back when the agent goes off track.
Beyond initial prompting, the tutorial covers ongoing workflow habits: answering the agent's clarifying questions thoughtfully, using project knowledge to reinforce design guidelines, and using the select tool for targeted edits. It also demonstrates using image generation and screenshots as design inspiration, and using version history to roll back when the agent goes off track. The walkthrough is positioned as applicable whether a user is brand new to Bolt.new or looking to sharpen an existing workflow.
Key facts
- 01The tutorial uses building a two-sided local marketplace for yoga classes as its working example.
- 02The enhance prompt feature rewrites and restructures a user's input for better agent comprehension.
- 03Plan mode causes the agent to generate a detailed implementation plan and ask clarifying questions before building.
- 04The presenter recommends using a performance model like Opus 4.7 for greenfield projects and smaller models for minor tweaks.
- 05Prompting incrementally — part by part — is advised over giving the agent too much at once.
- 06Additional tools covered include the select tool, project knowledge, image generation, screenshot inspiration, and version history rollback.
- 07Bolt.new runs entirely in the browser using WebContainers, with no setup or servers required.