Think Before You Prompt

Bjorn opened with a reframe that set the tone for the entire talk. Instead of just throwing a prompt into a tool, he argued, you should do what you would do in any real-world project: think about what you need and what the output should look like before you start building. That sounds obvious, yet the default behavior for most people using AI is to skip this step entirely.

He introduced a simple mental model: imagine a freelancer starting on day one. A junior designer walks into your office with zero context about the brand, the product, or the audience. What information does that freelancer need in order to deliver something great? That question becomes the blueprint for building your AI knowledge base.

“If a new junior designer starts on day one and has no idea, what information does that freelancer need to have in order to deliver something great?”
- Bjorn Vuylsteker

The freelancer analogy works because it forces you to externalize tacit knowledge. Every brand has unwritten rules about tone, layout preferences, color usage, and component structure. If you cannot articulate those rules to a person, you certainly cannot articulate them to a language model. The knowledge base is the mechanism for closing that gap.

Capturing Context: Screenshots, Voice Notes, and Specs

Bjorn walked through the practical tools he uses to capture raw context. The process starts with screenshots. He demonstrated taking a screenshot of an existing design, dropping it into Google AI Studio with Gemini Pro, and asking the model to write out the specs so a designer could fully understand the branding, color palette, typography, and card structures. The result is a structured description that becomes part of the knowledge base.

lightbulbScreenshot-to-Spec Workflow

Take a screenshot of any existing design or reference. Upload it to a multimodal model like Gemini Pro. Ask it to extract the visual specs: colors, typography, spacing, component patterns, and layout rules. Save the output as a Markdown file in your knowledge base.

For capturing thoughts and ideas quickly, Bjorn relies on Flow, a Whisper-based voice transcription tool. Rather than typing out detailed notes, he records voice memos that get transcribed automatically. For meeting transcripts specifically, he uses MeetMemo.app, which records and transcribes meetings automatically so every insight from user interviews, stakeholder calls, and team syncs flows straight into the knowledge base. The Product Consortium members get lifetime access to MeetMemo as part of their membership. The key habit is capturing first and organizing later, so nothing gets lost between having the thought and sitting down to build.

He described this collection of artifacts as a "Second Brain" for AI interactions. Voice notes, meeting transcripts, brand guidelines, competitor analyses, and product requirements all live as Markdown files in a structured folder. When it is time to prompt, you pull from the knowledge base instead of trying to reconstruct everything from memory.

The Context Framework: Five Pillars of a Good Brief

Bjorn presented a framework for structuring context that applies whether you are prompting a code generation tool, a design assistant, or a writing model. Five elements make up a complete brief that consistently produces high-quality output.

Who you are: your role, your expertise level, and the perspective you bring to the project
What you are building: the product, feature, or deliverable, described with enough specificity that the model understands scope
The user: who will use this, what their goals are, and what constraints they face
Constraints: technical limitations, brand guidelines, deadlines, budget, or platform requirements
Success criteria: what 'done' looks like, including acceptance criteria, quality benchmarks, and edge cases to handle

The framework is deliberately simple because the goal is adoption, not perfection. Even filling in three of the five pillars before prompting will produce dramatically better results than a bare request. Bjorn emphasized that models today follow instructions closely. If you provide a general prompt, the model will make assumptions about everything. But if you give it a list of components and the content, it will mimic exactly what you have written down.

infoSpecificity Drives Quality

Models today follow instructions closely. A general prompt leads to generic output because the model fills in every gap with assumptions. A detailed brief with components, content, and constraints produces output that matches your intent almost exactly.

Building a Second Brain for AI

The Second Brain concept ties together all of the capture methods into a sustainable system. Bjorn described how he maintains a collection of Markdown files organized by domain: brand guidelines, product specs, user personas, technical architecture, and meeting notes. Each file is written so that it can be dropped into a prompt context window with minimal editing.

The format matters. Markdown works because every major AI tool can parse it natively. It is lightweight, version-controllable, and human-readable. A brand guidelines file might include hex codes, font stacks, spacing scales, and component patterns. A product spec might include user stories, acceptance criteria, and data models. The discipline is keeping these files current and using them as the default starting point for every AI interaction.

Prompting Without Context vs. With a Knowledge Base

Without context, you prompt: 'Build me a landing page.' The model picks a random color scheme, invents copy, and structures the page based on generic patterns. With a knowledge base, you attach your brand guidelines, your copy document, and your component library spec. The model produces a page that matches your visual identity, uses your actual messaging, and follows the component patterns your team already uses. The difference is not a better model. It is better input.

check_circleCreate a dedicated folder for AI context files in Markdown format
check_circleAdd brand guidelines: colors, typography, spacing, tone of voice
check_circleDocument component patterns with examples and usage rules
check_circleCapture meeting transcripts and voice notes as raw material
check_circleWrite product specs with explicit acceptance criteria

The Prompt Is the Brief, the Context Is Everything

Bjorn closed by returning to the central idea of his talk: the prompt is just the brief. It is the short instruction that kicks off the work. But the context you provide alongside that brief is what determines quality, consistency, and relevance. A one-line prompt backed by a rich knowledge base will outperform a carefully crafted prompt with no context every time.

“The prompt is the brief. The context is everything.”
- Bjorn Vuylsteker

This shift in mindset changes how you invest your time. Instead of spending 30 minutes perfecting a single prompt, spend that time building and maintaining your knowledge base. The return compounds across every future interaction. Every new project, every new prompt, and every new tool benefits from the same underlying context. That is the real leverage of AI-assisted work: not the speed of generation, but the depth of understanding you give the model before it starts.

lightbulbContext Compounds

Time invested in building your knowledge base pays off across every future AI interaction. A brand guidelines file written once improves output for months. A product spec updated weekly keeps every generated artifact aligned with reality. The compounding effect makes knowledge base maintenance one of the highest-leverage activities for any product team using AI tools.