Through the Looking Glass
Three AI prompts that will tell you more about yourself than you meant to share
The interesting thing about a mirror is not what it shows you. It is what you do once you have seen your reflection.
A few weeks ago I posted a prompt called “the Cognitive Atlas”. It was a nice little attempt to turn an AI chat session into something closer to a self-portrait. The goal was to feed it everything I had written, everything I had prompted, the corners of my drafts and the questions I kept asking, and to ask it to render the shape of how I think back to me as an artifact rather than a paragraph (In this case a personality assessment á la Meyers-Briggs). People loved it! The comments and messages I got on Substack were fantastic: not “great post” but “this was uncomfortably accurate,” and occasionally “I deleted the output because I did not want it on my hard drive.” My favorite: “How I talk to Claude is not how I talk to my spouse of family.” Touché!
Still, I thought I’d follow up with a series of metaprompts that capture the self through the eyes of the machines we increasingly entrust with our deepest and darkest insights.
The Looking Glass
The current generation of consumer AI is in a scary way a mirror you cannot quite see the edges of. You bring it your calendar, your transcripts, your half-formed memos, your books-I-meant-to-read list, your prompts at 2am when you could not sleep. It returns some semblance of coherence and aids your productivity. The risk is that the coherence is built from a myriad digital corpus of you that has accumulated across hundreds of sessions and a dozen connectors, and the entity holding the mirror is a company whose business model is inherently profit driven. I just say this to flag the reality that the prompts below are fun. I think they are useful and creative. But the depth they reveal also sheds light on just how much of our privacy we are yielding to these AI models over time. So, with that disclaimer, carry on…
The Cognitive Persona
The first reflection. Get your own cognitive persona from Claude.
If you’re an AI super user (or not), input the prompt below and get your own personalized interactive analysis of your personality, prompting methodology, and cognitive style.
Build me a Cognitive Atlas — an interactive HTML artifact that maps how I actually think, where my wiring earns its keep, and where it costs me something. Not a personality test. Not a report. An instrument.
# The visual thesis
Cognitive instrumentation. Reference grammar: scientific atlases, Edward Tufte, Massimo Vignelli’s transit diagrams, Apple Health, Linear. The artifact should feel like a beautifully designed instrument panel for the self — dials, traces, polygons, towers, clocks. Prose is supporting material; visualizations carry the concepts.
# Phase 1 — Gather signal
Use whatever you have. If you have memory or prior context with me, mine it. If you don’t, ask compactly (one message, not a drip) for:
1. Voice sample. Two or three things I’ve written recently — emails, memos, posts.
2. Three decisions. Non-trivial choices from the past 12 months — what I weighted, what I declined, what surprised me about my own reasoning.
3. The work itself. What I build, what I keep coming back to, what I quietly refuse to do.
4. The shadow. Feedback I’ve heard repeatedly that I haven’t yet acted on.
Don’t begin Phase 2 until the signal is sufficient.
# Phase 2 — Build the artifact
A single self-contained HTML file. Stripe Press × Linear × Apple Health quality bar. Warm paper background. Fraunces serif for display, Inter for body, JetBrains Mono for metadata. Forced light background (survive dark mode). Generous negative space, restrained color, one accent and two atmospheric support tones.
## Structural requirement: Explorer + Scroll hybrid
Two modes:
- Scroll for first read: a linear narrative through diagnosis → practice → ritual.
- Explorer for return visits: persistent left rail (280px) with section anchors, animated active-section indicator, visited markers (small dots that fill in), and a Practice Mode toggle that filters down to only the actionable sections (vii–ix).
Rail collapses to a slide-out drawer below 1100px. Keyboard nav: jk or arrows between sections, p toggles Practice Mode, additional arrow keys for in-section interactions.
## Required visualizations — each section needs one
i. Hero. Two-word archetype name + an original SVG orbital diagram showing the core structure of my mind. Title words rise in sequence. Atmospheric gradient drift in 22–28s cycle.
ii. Type Inference. A vertical function stack tower showing dominant → inferior with a colored pulse that moves down the energy line connecting the four stages, then loops back up. Dominant stage washed in accent, inferior in shadow color, middle stages neutral. Hover lifts each stage.
iii. Big Five Trait Map. A hexagonal radar polygon that scales in from center on scroll, alongside detailed bars with behavioral evidence. Six axes: Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism, plus HEXACO Honesty-Humility. The polygon should read as a signature shape.
iv. Mental Model. An interactive braid diagram. 3–5 lenses positioned around a core (”decision crystallizes”). Threads draw themselves on load (staggered 200ms each). Hovering any lens brightens its threads. Each lens has a small three-bar weight indicator showing relative importance.
v. Fingerprint. Six patterns, each with a unique EKG-style signature trace in monospace stroke that animates in on scroll. The trace should encode the pattern’s character visually (rising for growth patterns, oscillating for compression, square-wave for rule-based, etc.). Plus prose and a monospace evidence quote.
vi. Values Core. Three commitments, each with a small line-drawn instrument symbol in single-weight stroke (1.5pt). Symbols should be original, not generic icons — a filter funnel, a measurement device, a rails diagram. Render like exploded technical drawings.
vii. Tensions & Shadow. Four strength/shadow pairs, each presented as a dial that rocks slowly between positions in a 7-second breathing cycle. Strength on the left (mossy green), shadow on the right (burnt sienna), with the dial visualizing that the same trait moves between contexts.
viii. Practice Lab. Four protocols. Each protocol’s steps shown as a horizontal flow diagram — numbered circle nodes connected by hairlines, monospace step numbers, with the line accent-colored on hover. Tag each protocol with the shadow it calibrates. Ground in: implementation intentions (Gollwitzer), pre-mortem analysis (Klein), cognitive reframing (Beck), self-determination theory (Deci & Ryan).
ix. Operating Manual. Two columns: Do (six items, mossy green) and Don’t (six items, burnt sienna). This is the one section that should stay text-only — the value is the directness, visuals dilute it.
x. Weekly Ritual. A clickable seven-day clock face. Days positioned around a circular face (Monday at top, then clockwise). Clicking any day fades in the question with a contextual note explaining which shadow or function that day calibrates. Arrow keys cycle days when the section is in view. Center has a slow pulse suggesting the ritual itself. Plus an optional annual prompt below.
End with a brief methodology section that is honest about what is inferred vs. measured.
## Animation principles
Restraint is the entire game. Every animation must communicate something.
- Title words rise in sequence on hero load (~120ms apart).
- Atmospheric gradients drift in 22–28s cycles.
- The orbital has slow rotational drift (80–140s per rotation).
- Section reveals trigger on scroll with subtle Y-translate (~20px) over 900ms.
- Grids stagger in at 80–100ms intervals.
- Trait bars and radar polygon animate on scroll-into-view.
- EKG traces draw themselves with stroke-dashoffset animation.
- Dials breathe continuously (7-second cycle) in their tension cards.
- Function stack pulse moves down the tower in 6-second cycle.
- Hover states use 300–400ms eased transitions.
- Paper-grain texture via SVG noise at 3–5% opacity, multiply blend mode.
No parallax theatrics. No cursor trails. No spring physics. No bouncing.
## Hard constraints
- Forced light background explicitly set on html and body.
- Never tell me I’m a hero. Tell me the truth. If something is a real shadow, name it without sandpaper.
- No AI tells: no “what this actually means,” no “here’s the thing,” no rhetorical em-dashes for drama, no “not X, but Y” antithesis constructions.
- Evidence over assertion. Every claim about me linkable to something I told you or you observed.
- Each visual does one specific conceptual job. Decoration is forbidden.
- Mobile-respectful below 1100px: rail becomes drawer, grids stack, flow diagrams become vertical lists, clock and radar shrink to centered.
- Self-contained: one HTML file. Inline styles and scripts. External fonts from Google Fonts permitted.
Build it.
Mine was scary accurate—and useful. I’ve taken several Meyers-Briggs before and usually get ENTJ. Claude inferred INTJ 👀, plus gave me a ton of insight on my prompting style
What did you get? Did AI nail it or miss the mark? Read on for 4 more prompts and artifacts like this one.
A quick note— for most of these prompts, after it runs an initial query in Claude CoWork or ChatGPT (if that’s what you decide to use), it will ask you several questions. Just be ready to type responses or use WisprFlow to dictate them more quickly!
AI self-knowledge prompts
I tried to write these prompt as if the AI were a forensic accountant rather than a coach. Give it permission to be unkind. Constrain it against its tendency towards sycophancy. Tell it explicitly not to call you a hero. Insist that every claim about you be linkable back to the data you fed it. Make decoration forbidden. Make flattery a failure warning.
When you do this, the output sounds a little less like a horoscope and starts sounding like an immersive chart review.
The three additional prompts below are the artifacts I have had a lot of fun with, and I’m hoping to archive these and see over the months and years how they change. Each prompt should produce an interactive HTML artifact you can save, screenshot, ignore for a quarter, and re-run.
A note before you copy. These prompts work best in Claude Cowork (where I built them and where the HTML artifact rendering is native), but they will run in any frontier model that can produce a single self-contained HTML file. They are long on purpose. I tried to include specific instructions and also make them fairly immersive.
The series name is Looking Glass, after the Carroll book. Alice walks through and meets a more truthful version of the world. Maybe we get that here too? Or perhaps it’s an exercise in the indulgence of our narcissistic qualities. Either way they were a lot of fun for me, and I hope they are for you too. I keep mine in a folder and I’m building a meta “looking glass” that draws on my various reflection prompts over time. I will be sure to share!
The Calendar
The hours you spend are the most honest thing about you. People lie on surveys. It’s harder to lie to their calendar (for the most part).
I ran this on ninety days of mine and it surfaced three standing meetings I would never accept today, a parallel-duplicate recurring series I had been carrying for a year without noticing, and a "starvation map” of commitments I had named as priorities and then not put on the calendar at all.
Copy the prompt below into a fresh Claude or ChatGPT conversation. Have ninety days of calendar export ready — ICS, CSV, or a pasted week-view will all work (If your MCP is connected on Claude Chat/Claude cowork it will render best). Also have a short note on available what you would say your priorities are, in case the model asks for it. See the video preview of my output here:
Expensing your subscription
Most readers expense Techy Surgeon through a professional-development, CME, or continuing-education budget, the same line that covers journals and conference fees. If you’re part of a department or group, reach out before you subscribe, there are group rates and a short justification note I can provide to make the approval easy. Here’s a template you can adapt and send today:
Subject: Professional development request — Techy Surgeon subscription
Hi [Approver name],
I'd like to expense a subscription to Techy Surgeon, a publication covering
healthcare AI, value-based care, and clinical/health-policy developments
relevant to [our department / my role in ____].
What it is: a regular analytical newsletter that translates emerging policy
(CMS payment models, AI governance) and clinical-AI tools into operator-level
implications for clinicians and health-system leaders. It functions as ongoing
professional education in areas directly tied to my work in [clinical care /
operations / value-based care / informatics].
Why it's worth it: staying current on these shifts informs [decisions I make on
____ / patient care / our value-based-care and AI strategy]. The cost is
[$___/year], well below the price of a single conference registration or
journal subscription, and the cadence keeps the knowledge current year-round.
Cost: [$___ / year or month]
Category: Professional development / CME / continuing education
Link: [techysurgeon.com or your subscribe link]
For departments or groups, the publisher offers group rates and a written
justification summary on request, which I'm happy to obtain.
Thank you for considering it.
[Your name, title]Prompt:
