CHAPTERS
- 0:00 – 2:35
Why prompts fail: unrealistic expectations and missing context
X.K. explains why a single prompt rarely produces great results: the model doesn’t know your goals, progress, or constraints. He frames the problem as both technical (model limits) and human (expectation management).
- •One-shot prompting assumes impossible mind-reading from the model
- •Prompt tweaking is often a symptom of insufficient background, not just bad wording
- •Two-sided issue: model capability + user expectations
- •Better outcomes require accumulating context over time
- 2:35 – 5:21
Context engineering as a workflow: accumulate materials and express intent beyond text
Kuse’s philosophy is to shift users from one-off chats to a compounding context system. X.K. argues that users should centralize files and artifacts and use visual/spatial tools to communicate intent with less verbal prompting.
- •Persuade users: stop relying on ‘one perfect prompt’
- •Accumulate documents and project artifacts in one place for compounding results
- •Use multiple modalities (spatial layout, selection, grouping) to express intent
- •Goal: better results with less effort as the system learns your project context
- 5:21 – 7:12
Inside the $10M ARR in 60 days story: long build-up, then explosive launch
Aakash challenges the ‘overnight success’ narrative; X.K. clarifies they built quietly since early 2024. The rapid ARR growth came after groundwork in product, distribution, and relationships—especially in local communities.
- •The ‘60 days’ headline hides months of prior product development
- •They built attention and connections before public marketing
- •Early traction came from Hong Kong/Taiwan communities (e.g., teachers)
- •Kuse 2.0 drove major spikes in traffic and demo requests
- 7:12 – 8:55
Hidden pivot: from design agent to document/knowledge workflows
Kuse began as an infinite-canvas design agent concept. User behavior revealed a stronger pull toward uploading and analyzing documents, so the team doubled down on knowledge and file-based use cases instead of pure design generation.
- •Original idea: design agent producing posters/design outputs
- •Users primarily used it as a file+document analysis workspace
- •Image model limitations and expectation gaps reduced design satisfaction
- •Late 2024 pivot: emphasize document understanding and knowledge workflows
- 8:55 – 12:26
Threads growth strategy: underserved channel + underserved geography + intern ‘content army’
X.K. details a distribution playbook centered on Threads, especially effective in Taiwan and Hong Kong. With no formal ads platform, they scaled organic reach through many accounts producing daily use-case content.
- •Threads offered unusually generous organic distribution vs. X/Twitter
- •Taiwan/Hong Kong markets were less competitive for AI apps
- •Scaled via many accounts and an intern team producing frequent content
- •Content focused on concrete, repeatable use cases (e.g., Formatter, exam papers)
- 12:26 – 17:01
Threads accounts demo + channel strategy: why Threads/IG beat X for user acquisition
They show the breadth of their Threads presence and discuss conversion realities. X.K. contrasts Threads’ discovery dynamics with X’s rigid creator hierarchy and recommends exploring Threads and Instagram for organic growth.
- •Demo of many Kuse-owned Threads accounts and example posts
- •Threads discovery is less hierarchical; easier for new projects to get reach
- •X is useful for VC visibility but hard for organic user acquisition
- •Instagram also works well for promotion alongside Threads
- 17:01 – 20:05
Visual context engineering: a ‘marketing term’ for multimodal intent + reusable 2D workspace
X.K. defines visual context engineering as giving users non-text ways to specify intent and relationships. The canvas acts like a 2D operating system for files and generations, making context reusable compared to linear chat history.
- •Coined term to communicate the product succinctly
- •Users express intent via spatial relationships, grouping, sketches, selection
- •2D canvas makes uploaded files and outputs visible and reusable
- •Supports a create→store→reuse loop instead of one-off chat exchanges
- 20:05 – 21:40
The mom analogy: context creates a positive feedback loop of better outputs
To simplify why context matters, X.K. compares AI to a mom cooking for a child: knowing preferences and goals leads to better results. More context improves output quality and encourages continued use, compounding value over time.
- •Context = knowing preferences, goals, constraints, and purpose
- •Better context produces better ‘tailored’ outcomes
- •Improved results create a virtuous cycle of trust and continued use
- •Reinforces the ‘AI as capable colleague/family’ mental model
- 21:40 – 25:33
RAG vs fine-tuning vs prompt engineering—and Kuse’s async, file-first approach
They map context engineering components and clarify how common techniques differ. X.K. explains Kuse relies heavily on RAG and document/OCR processing, with an emphasis on pre-processing files asynchronously for faster reuse later.
- •Context engineering includes prompt engineering, RAG, state/memory, structured outputs
- •Fine-tuning is heavy; Kuse uses minimal fine-tuning
- •RAG is core, especially for document-heavy workflows
- •Async pre-processing: prepare files ahead of time so future queries are faster
- 25:33 – 29:55
MVO before MVP: validate model outputs before productizing AI features
X.K. introduces an internal approach: Minimal Viable Output (MVO) before Minimal Viable Product. Because model responses drive user value, they prioritize stabilizing outputs before investing in full product development.
- •Traditional flow (PRD→build→ship) breaks when outputs are unreliable
- •Start by iterating on prompts/RAG/context until outputs are viable
- •Only then invest in UI/productization around the capability
- •Reframes AI building as output engineering first
- 29:55 – 36:28
Demo: creating a PRD in Kuse using simple prompts + strong context
X.K. walks through Kuse’s three-step workflow: drop files, select content, ask. The demo shows generating a PRD from multiple documents and controlling behavior with tools like ‘Source only’ to ground answers in uploaded materials.
- •Workflow: drop files onto canvas → select → ask → get structured output
- •PRD generation example using several PDFs and a graphic as context
- •‘Source only’ constrains outputs to uploaded documents (reduces hallucination)
- •Model choice options (GPT/Claude/Gemini/DeepSeek) depending on task
- 36:28 – 38:22
Prototype generation + ‘don’t pretend’: practical value over ‘AI wrapper’ insecurity
They generate a simple prototype page from the PRD and discuss what happens behind the scenes (summarize, then hand off to a coding-capable model). X.K. argues users care about solved problems, not whether the product is ‘just a wrapper.’
- •Webpage generator creates prototype from existing PRD/context
- •Pipeline: summarize/layer context → send to model (e.g., Claude) for build
- •Honest positioning: straightforward solutions beat artificial complexity
- •Value comes from workflow + context system, not model mystique
- 38:22 – 44:43
Why Kuse vs Claude Code/Lovable/Bolt: compounding context and broader non-coder workflows
X.K. positions Kuse as context compounding infrastructure rather than ‘one prompt to build a product.’ He highlights different target users (non-coders, PMs, HR/admin) and describes how the same context workspace supports repeated iterations and sharing.
- •Kuse requires more context upfront but improves with reuse over time
- •Targets non-engineers; engineers may prefer Cursor/Claude Code
- •Use cases extend beyond prototyping (e.g., HR announcements, link hubs, internal pages)
- •Canvas can store outputs alongside inputs for iterative loops
- 44:43 – 57:18
Advice for AI founders: ignore FOMO, follow users, focus over competitors, and rethink ‘productivity’
X.K. shares founder guidance: don’t obsess over OpenAI updates or competitors like Miro/Dropbox; focus on users and iterate with what they actually do. He also reveals their funding story (self-funded via trading profits) and offers a philosophical view that ‘productivity tools’ may evolve into ‘playgrounds’ for meaning and enjoyment.
- •Don’t be paralyzed by platform risk or rapid model updates—focus on users
- •Follow real usage signals (their pivot was user-driven)
- •Competition framing shifts over time; stay anchored to value and mission
- •Self-funding enabled focus; VC fundraising can distort priorities
- •Long-term thesis: tools may shift from efficiency to fulfillment/entertainment
