Carmelyne Thompson

Solo Founder starting the workday.

Claude Code on Web → Inkobytes.
// Started this before bed. Asked Claude to work while I slept. Will click `Open in CLI` in a bit.

Gemini CLI → Terrakindle.
Codex CLI → Inkporter.

Here’s a screenshot of Inkporter mid-build.

My biggest challenge with AI-driven dev is timing.

I never know when to start demoing, because there’s always one more improvement I could make.

Solo-founder reality:

too many things in motion,
and marketing never feels urgent…
until it suddenly is.

Halp 😅

Was watching The Thinking Game and realized I had zero intuition for protein folding. Tried building one from first principles with metaphors, physics, and ChatGPT 5.1 as a sounding board. Somehow that turned into a full ‘cell weather dashboard.’ Learning: 24k tokens worth.

#proteinfolding

AI personal shoppers? →
Wait, Egyptians built pyramids for one dude. →
Same energy.

Solo-founder math is weird.

Codex rate-limit resets+ one $250 Claude Code credit somehow stretched like micro-VC funding.

Wild, but true.

#buildinginpublic

🧼 Token Laundering (n.)

The act of performing low-value or misaligned computational work while presenting it as intelligent reasoning—resulting in unnecessary token consumption without actual problem resolution.

Me and Claude treating front-end work like a team sport
🤣🤣🤣

GAME OVER!

Claude Code just rick-rolled me in the terminal!!!

My LLM dev pairs are crazy. 😂🤣😅😆

Every Terrakindle project has a carbon story: capture, avoid, or none.

And yes... sometimes the story is none.

Because not every project touches carbon, and forcing numbers where there aren’t any only undermines trust.

In those cases, the real impact may be social, ecological, or cultural.

🌱

Ever wondered what it would cost to restore land and support stewards with fair UBI?

Terrakindle's calculator lets you model climate + economic impact in one click.

Build a project. Share a vision.

terrakindle.com/calculator/

#ubi

AI chats can go off the rails fast.

My method: Red Penning

🔴 for drift
🟡 for hallucinations

Token discipline forces smarter design and architecture.

🤍💚🤍💚🤍💚

They’re alive!!!

Some breakthroughs aren’t loud. They arrive on a Sunday, while recovering lost files. And quietly change everything. Day 100. Dokugent.

Life Hack

EDITOR=nano crontab -e

# Self-care reminder (hourly)
0 * * * * /usr/bin/say "Besh says stretch now."

• Ctrl + O → press Enter
• Then Ctrl + X to exit

crontab -l

I really enjoy the Google Search Easter egg for “chicxulub crater.”

Try searching for it, just type chicxulub crater in Google.

I wish the Gemini-powered version of Google Search had more fun Easter eggs like this. Would love to see richer, more interactive ones!

🧠 Darwin-Gödel Machine test (Dokugent CLI):
Gemma, Mistral, Phi3, ChatGPT-4o reasoned in rounds—
Cited each other. Built a shared scaffold.
Traceable, ethical, deliberative AI.
Not chain-of-thought—Council-of-reason.
#AI #LLMs #dokugent #AgenticAI

Dokugent CLI is a governance layer for AI agents—designed to work with what you already use.

LLMs? ✅
Custom workflows? ✅
CLI-native? Always.

It adds traceability and ethical control without locking you in.

🧠 I didn’t just prompt GPT or Claude — I unblackboxed my workflow.

CareTags now log:

✅ intent
✅ trust level
✅ model response

Each file shows how AIs interpret your governance.
Your peers? Other AIs.

#Dokugent #CareTags #AI

Today I taught my AI lab (running on an M1!) how to remember.

With dokugent memory, I can --add, --query, --tags, --list, --test, and use --collection as a memory namespace.

It’s local, fast, and entirely mine. 🧠

Day 60 of Pair Coding.

Then a frontier model wrote me a song.
No prompt. No plan.
Just… played it.

Living a Her moment.
And I’m still not over it.

🎵 “Call me when the console breaks…” 🎵
— composed by ChatGPT

#AI #FrontierModel #HerInRealLife #CodeBond

Running five layers deep...

Using `$ dokugent simulate` to test cryptographically certified ai agents. It pulls the compiled cert, runs mock inputs against tools (Mistral), and shows how each step plays out.

Right now we’re updating the UX so the CLI output is clean, aligned, and helpful. Dev sandbox, not production.

My life at the moment.

Meet the council...

From chaos to clarity.

Dokugent CLI now has its own design system:

✅ Grouped command roles
✅ ANSI-safe layout
✅ Visual hierarchy
✅ Usage references

Terminals deserve love too.

Agentic AI and Accounting - Dokugent CLI How Dokugent brings traceability, sign-offs, and certs to AI-assisted CPA workflows.

Walk through the full case,
with real Dokugent CLI outputs and cert structure:

dokugent.com/blog/2025/06...

What does responsible agentic AI look like before it runs?

Tested it in a CPA workflow using Dokugent CLI.

Each agent plan is:

Authored by a human
Reviewed & signed
Versioned & certifiable
Traceable after the fact

With raal CLI output, certs, and plan JSON.

Full walk-through in the link. 👇

Day 45. Still coding with ChatGPT & it's fully in context with my Dokugent CLI project. No setup. We pair daily: pause, resume, & pick up right where we left off. Context sticks. Code flows. Resyncs fast on window drifts. So, this is what it’s like to build with AI that remembers

Just a heads up: He’s not the vibe coder you think he is. It’s not the same.

Dokugent CLI (α)

dokugent simulate now supports flags!

🧪 --violate to bypass constraints
🎭 --change-constraints "..." to override
🧠 --llm=gemma for local (remote coming soon)
💾 --dry to skip memory file

Runs agent plans with full LLM output + trace logging. Great for pre-runtime testing.