See CandleKeep in Action
Real results from production sessions. See how domain expertise transforms AI agent output.
Security Research
Your agent reads the security textbook so it audits code like a specialist, not a generalist.
The Problem
When you ask an AI agent to find vulnerabilities in a codebase, it gives you the OWASP Top 10 checklist and calls it a day. It misses injection variants, doesn't understand metacharacter attacks, and can't apply formal auditing methodologies.
What devs try today
Copy-paste security checklists into prompts. Run static analysis tools that produce hundreds of false positives. Or manually cross-reference 1000+ page security references while your agent gives you surface-level advice.
Imagine if...
Your agent opens "The Art of Software Security Assessment" to the chapter on metacharacter injection, reads the vulnerability patterns, and systematically audits your codebase — citing which chapter and technique flagged each finding.
Three findings based on metacharacter injection patterns: (1) The shell exec call on line 42 passes unsanitized user input — use execFile() with an argument array instead of string interpolation. (2) The template literal in the SQL query on line 87 is vulnerable to second-order injection via the username field. (3) The file path construction on line 123 doesn't normalize paths — vulnerable to directory traversal via ../ sequences.
“Metacharacters are characters that have special meaning to a program or protocol beyond their use as regular data.”
The Art of Software Security Assessment
Mark Dowd, John McDonald, Justin Schuh
The definitive 1,130-page reference on identifying and preventing software vulnerabilities
Business Strategy
Your agent synthesizes insights from 12 business books to craft strategy, not generic advice.
The Problem
AI agents give you MBA-textbook answers — "identify your target market" and "create a compelling value proposition." When you need specific persuasion frameworks, pricing psychology, or outreach techniques, the agent can't go deeper than blog-post-level summaries.
What devs try today
Copy-paste frameworks into prompts. Ask the agent to search the web (getting SEO-optimized blog posts instead of primary sources). Or read the books yourself and manually apply the frameworks.
Imagine if...
Your agent consults "Influence" for persuasion principles, "SPIN Selling" for question frameworks, and "$100M Offers" for pricing psychology — synthesizing insights from 12 books into one actionable strategy.
Use the SPIN framework: open with a Situation observation ('I noticed your team ships 3 releases/week'), bridge to a Problem ('most teams at that pace accumulate security debt'), then Implication ('one missed vulnerability at scale becomes a headline'). Close with a Need-payoff question, not a pitch. Per Cialdini's commitment principle, ask for a small yes first — 'Would a 2-minute case study be useful?' — before requesting a meeting.
“The purpose of Situation Questions is to establish a context... leading to a better understanding of the buyer's Problem.”
Influence: The Psychology of Persuasion
Robert Cialdini
The science behind why people say yes
SPIN Selling
Neil Rackham
Question-based framework for high-value sales conversations
$100M Offers
Alex Hormozi
How to structure offers so the decision is 'your product vs nothing'
UI/UX Design
Your agent applies design principles from 7 books, not generic 'make it pretty' advice.
The Problem
AI agents produce functional but generic UIs. They follow Tailwind defaults and component library patterns without understanding the cognitive psychology behind great interfaces — why certain layouts reduce cognitive load, how visual hierarchy guides attention, or when to break conventions.
What devs try today
Share design system docs in context. Paste screenshots and ask for improvements. Or accept the cookie-cutter output and hire a designer later to make it feel polished.
Imagine if...
Your agent consults "Refactoring UI" for spacing and hierarchy rules, "Laws of UX" for psychology-backed design principles, and "Don't Make Me Think" for usability patterns — producing interfaces that are both beautiful and cognitively optimal.
Apply Hick's Law: limit the top-level navigation to 5-7 items max — more options increase decision time exponentially. Use the F-pattern for the metric cards (primary KPI top-left, secondary top-right). Per Miller's Law, chunk the data table into groups of 4±1 columns with progressive disclosure for details. The action buttons should follow Fitts's Law — larger and closer to the cursor's resting position.
“The time it takes to make a decision increases with the number and complexity of choices available.”
Refactoring UI
Adam Wathan & Steve Schoger
Practical visual design techniques from the creators of Tailwind CSS
Laws of UX
Jon Yablonski
Psychology-backed principles for interface design
Don't Make Me Think
Steve Krug
The usability bible for web interfaces
Ready to make your agent a domain expert?
Start building your knowledge library today. Free to get started, no credit card required.
Get Started Free