Aakash Gupta$10M ARR in 60 days with context engineering
Aakash Gupta and Xiankun Wu on context engineering powers Kuse’s viral growth and rapid $10M ARR.
In this episode of Aakash Gupta, featuring Aakash Gupta and Xiankun Wu, $10M ARR in 60 days with context engineering explores context engineering powers Kuse’s viral growth and rapid $10M ARR Prompts often fail because users expect perfect results without giving sufficient project background, so repeated tweaking is an expectation-and-context problem as much as a model-capability problem.
At a glance
WHAT IT’S REALLY ABOUT
Context engineering powers Kuse’s viral growth and rapid $10M ARR
- Prompts often fail because users expect perfect results without giving sufficient project background, so repeated tweaking is an expectation-and-context problem as much as a model-capability problem.
- Kuse’s core thesis is “context engineering”: accumulating documents, state, and intent over time so AI behaves like a long-tenured colleague rather than a one-off chatbot session.
- The company’s rapid revenue narrative was enabled by long “silent” building plus a distribution wedge in Taiwan/Hong Kong via Threads, where an “intern army” scaled many use-case accounts into organic traffic.
- “Visual context engineering” is positioned as both a UI and workflow: a 2D canvas/whiteboard that makes it easier to organize, reuse, and combine files and AI outputs, reducing reliance on sophisticated prompting.
- For AI product development, X.K. emphasizes validating “minimal viable outputs” (MVO) before investing in full “minimal viable product” (MVP) builds, and staying focused on users rather than fear of platform shifts or competitors.
- He argues startups shouldn’t over-defend against the “AI wrapper” critique: users care about solved problems, and long-term value comes from compounding context rather than novelty features.
IDEAS WORTH REMEMBERING
7 ideasPrompting breaks when context is missing, not when users lack “magic words.”
X.K. compares prompting to hiring a new employee: with no background on goals, constraints, and progress, perfect execution from a short request is unrealistic, so users end up iterating endlessly.
Design products to accumulate context so results improve with usage.
Kuse pushes users to store materials in one place and reuse them, creating a compounding loop where the system knows more about the project over time and needs less prompting.
Use multiple intent channels (visual + selection + structure), not just text prompts.
“Visual context engineering” frames the canvas as a way to express spatial relationships among docs/objects and to select/recombine inputs, making intent clearer than pure conversational chat.
RAG is the workhorse for doc-centric products; fine-tuning is optional and heavy.
Kuse relies heavily on RAG plus strong file/OCR/document processing; X.K. downplays fine-tuning as resource-intensive relative to the product’s primary needs.
Preprocess files asynchronously to make downstream AI interactions faster and smoother.
Instead of doing all retrieval processing at query-time like many chatbots, Kuse processes folders/documents ahead of time so future tasks feel like “ingredients already on the table.”
Validate “minimal viable outputs” before building the full feature experience.
Their internal MVO concept prioritizes stabilizing model responses first; only once outputs are good do they invest in productization, which reduces wasted engineering on brittle AI behaviors.
Distribution wedges can be geographic and platform-specific, not just product-led.
Kuse leaned into Taiwan/Hong Kong (less crowded) and Threads (generous reach, weaker creator hierarchy) while scaling content via many accounts run by interns to create daily use cases.
WORDS WORTH SAVING
5 quotesPeople expect AI can deliver exactly as people wish within such a short description… is basically impossible.
— Xiankun Wu
Context engineering is like your mom knows you very much… so she can cook something that caters to your purpose.
— Xiankun Wu
Before you have the minimal viable product, you should have… minimal viable output first.
— Xiankun Wu
If it is the useful solution, don’t pretend to be… creating a very complicated… solution here.
— Xiankun Wu
Entrepreneurship is a game of focus.
— Xiankun Wu
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsWhat exactly counts as “context” in Kuse—project goals, constraints, prior outputs, brand voice, user preferences—and how is it represented internally?
Prompts often fail because users expect perfect results without giving sufficient project background, so repeated tweaking is an expectation-and-context problem as much as a model-capability problem.
In the canvas UI, what specific visual signals (position, grouping, arrows, proximity) change model behavior, versus being only organizational for the user?
Kuse’s core thesis is “context engineering”: accumulating documents, state, and intent over time so AI behaves like a long-tenured colleague rather than a one-off chatbot session.
You emphasize heavy RAG: what retrieval strategy and chunking approach works best for mixed PDFs, images, and OCR text in real customer folders?
The company’s rapid revenue narrative was enabled by long “silent” building plus a distribution wedge in Taiwan/Hong Kong via Threads, where an “intern army” scaled many use-case accounts into organic traffic.
How do you prevent “context bloat” (too many files) from degrading answer quality or increasing latency/cost, and what heuristics decide what to include per task?
“Visual context engineering” is positioned as both a UI and workflow: a 2D canvas/whiteboard that makes it easier to organize, reuse, and combine files and AI outputs, reducing reliance on sophisticated prompting.
Can you break down the Threads “intern army” playbook—account count, posting cadence, content templates, and the funnel from impressions to demos to paid?
For AI product development, X.K. emphasizes validating “minimal viable outputs” (MVO) before investing in full “minimal viable product” (MVP) builds, and staying focused on users rather than fear of platform shifts or competitors.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome