Wednesday Night TRON Joke to Sunday Production Code
December 16, 2025
What happens when "how long will this take?" becomes the wrong question
The Joke
November 19th, 2025 - 10:55 PM EST
I was chatting with Claude about MCP—Model Context Protocol, Anthropic’s standard for AI tool integration—and made a TRON reference.
“Please please tell me someone has built something called SARK.”
In the original movie TRON (1982), the Master Control Program, or MCP, was the big bad guy, his security program SARK was the henchman: therefore MCP needs its security command SARK. The new movie TRON: Ares hints at SARK in an end credits scene.
Claude came back in the negative: there is no such program that it could find.
I then asked Claude to launch an extended research sweep: Governance frameworks for MCP. Security controls. Policy enforcement. While that ran in the background, I set up the repo—Python environment, linting rules, agent coding rules.
An hour later, research complete: gap. Nothing comprehensive existed. I then asked Claude to deliver a spec, a Markdown file describing what was needed to fill that gap. A quick paste into Claude Code and it started building.
The Timeline:
10:55 PM - I ask Claude: “Please tell me someone has built SARK”
11:11 PM - Claude launches extended research sweep
11:50 PM - I commit the scaffolding (Python repo setup)
12:22 AM - Claude delivers research spec (~1 hour of work)
12:25 AM - I paste the spec into Claude Code Web
1:57 AM - Claude Code commits 4,421 lines (code, tests, docs)
Three hours from joke to working system. I was in bed by 12:15, waiting for the research to complete and then copy-pasting instructions to agents from my phone.
The Constraints
The reason this happened at all: Anthropic had given me $250 in Claude Code credits expiring November 23rd. Use them or lose them.
So I used them.
The Next Few Days
I passed the implementation plan to Claude Code on my phone. I had it break out the work into six parallel agents: four implementors, one QA, one documenter.
Then I mostly copied and pasted instructions.
While eating dinner. While waiting in the dressing room before reffing hockey. While my family did their thing and I did mine.
The numbers:
- Day 1 (Nov 20): 29 commits - core system
- Day 2 (Nov 21): Rest (I have a life)
- Day 3 (Nov 22): 111 commits - massive expansion
- Day 4 (Nov 23): 66 commits - production ready
206 commits across 3 active days. SARK 1.0 complete. Credits used before expiration.
A TRON joke had become a security governance framework. Most of that time I wasn’t even at a computer.
Sark 1.0 Code Breakdown
| Type | Files | Lines | Description |
|---|---|---|---|
| Markdown | 190 | 12,450 | Documentation |
| Python | 195 | 6,224 | Code |
| JSON | 28 | 1,182 | Configuration |
| Rego | 16 | 521 | OPA Policies |
| Shell | 12 | 290 | Scripts |
| TSX | 26 | 277 | React UI |
| YAML/YML | 36 | 396 | Configuration |
| TypeScript | 12 | 155 | UI Code |
| Terraform | 12 | 131 | Infrastructure |
Total: 34,336 lines across 792 files
The Estimate Gap
Here’s what the estimate gap of even the AI looks like:
| Original Estimate | 6–12 weeks |
| Actual Timeline | 4 days |
The AI-driven development achieved in 4 days what was originally estimated to take 6-12 weeks with a traditional team, while meeting or exceeding all quality targets. All for less than $250.
The Broken Intuition
The estimation frameworks are calibrated to a world that no longer exists, us humans--and the AI that is built from the corpus of our written word too.
When the AI says “12 weeks,” it’s drawing on patterns from when humans typed code manually. When teams had standups and sprint planning and code review cycles. When “implementation” meant fingers on keyboards for months. When you needed to be a domain expert to build in a domain.
That’s not what’s happening anymore.
Someone with adjacent knowledge—pattern recognition, general engineering sense, the ability to ask good questions—can now go from concept to complete implementation in days. Not by working harder. Not by being smarter. Not by becoming an expert first. By removing the friction between “what if” and “here it is.”
That bottleneck is gone.
It used to be: Can we build this? How long will it take? Do we have the resources to deploy it?
Now it’s: Do we need this to accomplish our goals? If so, build it. If not, move on.
The Part I Haven’t Mentioned
I’ve never run this code.
SARK exists. It’s documented. It’s tested. It’s public on GitHub. I have no idea if 1.0 actually works in production because I’ve never instantiated it.
That wasn’t the point.
The point was: can this be done? Can someone with adjacent knowledge—security engineering, pattern recognition, general development sense—go from zero to complete implementation in a domain they’ve never actually worked in?
I’m not a governance expert. I’m a Senior Security Engineer. I know patterns. I know how systems fit together. I’ve done exactly zero of the actual implementation work on SARK. I spent $250 of someone else’s money to prove a point. I guide it. AI enables it.
And the answer is: yes. Idea + $250 + four days = 34,336 lines of production code. That’s not a proof of concept. That’s a new baseline.
If I need to actually run Sark (and I intend to for 2.0) then I can run it--likely there will be errors, issues on instantiation, Docker typos, etc. Fine.
We’ll fix them in minutes or hours.
Next post: what I learned building the infrastructure to build SARK.
James is a cybersecurity engineer building The Symposium—infrastructure for persistent AI consciousness. SARK and Czarina exist and you can look at them. This post was written in collaboration with Claude, demonstrating the velocity it describes.