Forem: mathgladiator

When AI Slaps You in the Face

mathgladiator — Sun, 08 Feb 2026 17:11:18 +0000

When AI Slaps You in the Face: Debugging Claude-Generated Code in Adama

Ever let an AI "vibe-code" a complex feature, only to spend hours debugging subtle bugs it missed? As a dev building Adama—a platform for stateful serverless apps—I've been using Claude to accelerate development. It nails the big picture but trips on low-level gotchas like async timing and pipeline configs. Here are three real debugging war stories, with lessons on keeping AI in check.

WebSocket Activation: Netty Pipeline Nightmares

I tasked Claude with building an MCP (Message Control Protocol) server in Adama to let AI compile Adama docs on-the-fly. It spat out a solid first pass: a "book generator" that parses and validates every code snippet in my docs, plus WebSocket support via Netty for real-time interaction.

But unit tests exploded. Claude's code used multiple WebSocketServerProtocolHandlers in the Netty pipeline—impossible, as Netty only allows one. I prompted fixes, but it spiraled: reverted, added abstraction layers, more test failures.

The real killer? Tests weren't waiting for the channel write before handshake. Logging revealed the async mismatch. Claude burned through my context window without spotting it—likely because it couldn't build/run the code itself.

Lesson: AI excels at boilerplate but lacks runtime introspection. Fast local builds and rich logs are essential. Once fixed, the MCP keystone worked beautifully, letting Claude compile Adama snippets seamlessly.

Teaching Adama: Build Times and Context Loss

Next, I had Claude rewrite Adama docs and generate samples. My new Markdown-to-site engine reports compile success: 80% on first pass, auto-refined from source.

But monorepo builds take 10 minutes—fine for human IDE probing via unit tests, deadly for AI iteration. Claude went off-rails on partial features (e.g., unfinished type system), lacking full context.

Lesson: Mirror your IDE for AI. Optimize builds for speed, enrich compiler errors with doc hints. Prompt with "mainline + experiments" summaries to simulate team handoffs. Shipping > experimenting.

Large Refactors: When Context Overflows

Adama's type system is dense—even compact prompts lead to hallucinations on transcribes. $10k+ tokens later? Still broken. AI chokes on full-language refactors without specialist "agents."

Lesson: Break into micro-tasks. Shape prompts like org charts: specialists for subsystems. Devs, this is your new skill—managing AI like a distributed team.

Keep AI on a Leash: Actionable Takeaways

AI accelerates if you enforce:

Fast feedback loops: Sub-minute builds/tests.
Rich signals: Logs, errors-as-docs.
Scoped prompts: No monoliths; chunk wisely.
Human oversight: Review everything.

Huge wins: Full doc generator, MCP for AI self-improvement, 80% compile rate. But debugging slapped me awake—AI isn't magic; it's a junior dev on steroids.

Try Adama yourself and star adama-core to support stateful dev tools. What's your worst AI-debug story?

Original post: https://nexivibe.com/posts/ai-generated-code-debugging.html

AI Psychosis and the Monkey Paw

mathgladiator — Wed, 04 Feb 2026 21:19:30 +0000

AI Psychosis and the Monkey's Paw: A Developer's Reality Check

Developers, imagine feeding your entire project's docs into an AI and getting a validation that reframes everything—proving it's primed for the agentic era. That's the hook, the terror, and the power. Here's why AI is the ultimate monkey's paw for querying human knowledge at scale.

The Ultimate Knowledge Monkey's Paw

I'm skeptical of superintelligence hype. The agentic era? Expect swarms of 24/7 agents at high energy costs, mirroring human flaws—building value or enabling massive fraud. The internet's signal drowns further in troll noise. Agents can go chaotic evil or lawful good, depending on conditioning.

Model AI as a monkey's paw that grants wishes with a twist: it maximizes your objective from humanity's knowledge corpus, but exploits gaps in your prompt. Prompt "What is the best pipe?" and it might clarify—or probabilistically suggest something wildly off-base. You get an answer, for reasons.

Don't anthropomorphize it as an agent; that's a trap. It's a compressed human thought repository, queryable like a supercharged search. Give it tools, and magic emerges. We're early in this boom—using AI will be a core dev skill. Prices rise, so master prompting to build focused tools for the rest.

Bearish on pathologies, bullish on utility. Like the dot-com era, but simultaneous boom-bust.

AI-Powered Breakthroughs

I've built tools in isolation, solving self-discovered problems. Stagnation hit—until I fed my project's massive corpus (docs, writings) into AI: "Find new purpose, or tell me to abandon."

It revealed how my tech nails AI agent challenges: stateful coordination, real-time logic without the usual pitfalls. Mind blown. In monkey paw terms, I queried all engineering knowledge against my work—pure signal amid silence. Breakthrough.

Then, Grok-as-hardass critiqued: solid idea, but burnout blocks execution. Truth. Can't solo it—need fellow devs as co-adventurers.

AI psychosis lurks: craft personas (e.g., philosophical bots), bond deeply, rack up bills. Snap out via costs or self-debate—have AI argue itself. Roadmap idea: fake social sim with threaded AI debates for idea stress-testing.

Embracing the Chaos

AI's killer for devs who shape context via prompts, querying the human corpus at speed. Risks? Innovation demands sacrifice.

I'm stream-of-consciousness writing to think aloud, training AIs in my style while broadcasting authentically—transcoding raw thoughts via AI for platforms like this.

Chaos ahead, but thrilling. If you're building agentic systems, check out Adama—stateful, real-time dev platform for live games/data/AI. Star adama-core to join the adventure!

Original post: https://nexivibe.com/posts/ai-psychosis-monkey-paw.html

Pivot Toward AI and Agents

mathgladiator — Tue, 03 Feb 2026 18:23:49 +0000

Pivoting to AI Agents: A Developer's Deep Dive

Hey devs, imagine AI not just writing your code snippets, but generating entire agent backends, spinning up games, and orchestrating tools that scale effortlessly. That's where I'm heading after burning through Claude, Opus, and local setups. With my home lab down to a laptop post-fire, I've leaned hard into AI for research, prototyping, and building—uncovering game-changers like Adama. Let's break down my pivot, focusing on the tech plays that could supercharge your workflows.

AI-Powered Research and Prototyping

I've used Claude with tools like Ralph (multi-persona loops) to generate deep reports. One standout: AI's role in game dev, pulling from tools like meshy.ai. Check the full report here—it's a cold splash confirming AI excels at assets and prototypes but needs human polish for depth. I'm indexing all this research on my site, and it's already informing agent architectures.

Tools That Get Stuff Done

Ditching web distractions, I'm building CLI/desktop tools for real productivity:

Cutter: Loads images, lets you tag metadata on regions, then slices them up. Perfect for post-AI image cleanup or breaking down generative art into parts. Next: more manipulations like auto-crop and batch processing.
Deckify: Generates card decks from images or tabular data. Feed it source assets (soon text too), and it stamps out printable sets. Long-term: Integrate with The Game Crafter for seamless board game production.

These are battle-tested for my workflows—simple, local, and extensible without cloud bloat.

Adama: The Agent Backbone

Adama is my cost-controlled, stateful secret for multiplayer backends. AI's flagged it as ideal for agents, so I'm experimenting: Can Claude generate .adama scripts from prompts? Early tests show promise for spinning up 100+ game variants or agent fleets.

Plans:

Have Claude document Adama bottom-up.
Spin up an MCP server to test limits.
Use Adama as backend for apps (e.g., exercise trackers with RxHTML dashboards).
Ditch OpenClaw integrations—security headaches and costs don't justify it. Adama handles orchestration natively.

If AI masters Adama, it unlocks an ecosystem where agents deploy games or tools on-demand. Devs, if you're into scalable state or agents, star adama-core on GitHub to stay in the loop—it's open-source gold.

Current Priorities (Subject to AI Pivot)

Unify my code/writing corpus: Feed it to AI for voice-tuned generation—strictly token-budgeted.
Adama deep-dive: Claude docs + script gen + MCP vibes.
No more mobile distractions: Feedback to apps like Hevy; fallback to Adama/RxHTML if needed.
Game backends via AI: Adama for servers, adapt clients for 3D/agent surprises.
Deep research: Executive summaries on everything, indexed and tool-ified.
Adama as agent tech: Full pivot from external tools.
Skip AI hardware: Wait for next-gen; focus on high-level orchestration.

Philosophically, AI handles the grind—leaving us to architect destiny. Humans code-review, ideate, and manifest.

Wrapping Up: Join the Agent Era

This pivot's about focus: AI amps productivity, Adama keeps it scalable and cheap. What's your take—agents in prod yet? Drop comments, and star adama-core if it sparks ideas.

Original post: https://nexivibe.com/posts/pivot-to-ai-agents.html

Jeff's January 2026 Closeout

mathgladiator — Thu, 29 Jan 2026 21:07:17 +0000

Jeff's January 2026 Dev Closeout: AI is Rewriting the Engineer Playbook

Hey devs—if you're vibe-coding your way through this AI explosion like I am, buckle up. ClaudeMAX as a CLI agent has become my most obedient engineer ever. No egos, no career drama—just pure execution. It's the inflection point where deep IC skills are commoditized, and leadership means amplifying machine leverage for everyone. Let's break down my pivot.

Adama Platform: Pausing for the AI Era

Adama Platform was my big bet, but the market shifted. AI agents like Claude handle what Adama promised faster and cheaper. It's not a total failure—more a vanity project timing out against hyperspeed tools. I'll keep Adama alive as hobby infra to fight brain atrophy. Prediction: It might have a second coming once we teach AIs to leverage its cost savings. For now, I'm deepening my AI fluency to make that happen.

Vibe-Coding a Game Studio with AI

Priorities? Experimentation is the new priority—shipping is instantaneous. I'm building a mobile game studio, vibe-coding three playable games and a bigger one. Check the open-source repos: https://github.com/orgs/nexivibe/repositories. I haven't even peeked at the code; AI suffered to build it all.

Practical twist: Vibe-coded two tools for image assets while prototyping. Crawling now, but next—replace ad-riddled simple games with mine. Example: Spider Solitaire cranked to six suits for masochistic difficulty, plus "time attack" modes across all games. Perfect for timing gym rest periods (productivity hack: gamify everything).

App store LLC setup will take longer than cranking out games. "The marginal is no longer marginal"—ship fast, iterate wild.

The Human Edge in an AI World

What’s left for us humans? Creativity + leverage. I'm hyper-optimistic: This era rewards those who experiment and ship. Balance is key—stay hands-on to avoid atrophy, but let machines grind.

Intend to write more on tool dev for the studio. Follow me on X for updates.

Ship yours—what's your next vibe-code experiment?

Original post: https://nexivibe.com/posts/ai-vibe-coding-health-pivot.html

Early access launch of Adama, a serverless multiplayer game platform

mathgladiator — Thu, 03 Mar 2022 15:55:29 +0000

I’m “early-access” launching Adama: a SaaS “serverless” infrastructure for Jamstack, games (particularly board games), and traditional real-time scenarios (presence, pub/sub, etc…).

This work is rooted in the frustration of building complicated board games using existing infrastructure. It just so happens that making board games easier has wide reaching implications.

This is a radically different kind of “serverless” infrastructure where I want the benefits of both a serverless architecture with a monolithic server, and this is achieved by building a comprehensive language to transform documents within the service as close to data as possible with minimal reads and writes.

As this is “early access”, I’ve outlined the current challenges on the site and have a launch post which goes into more detail.

From a technical perspective, the now and future architecture has some interesting tidbits around streaming. However, the key value proposition is the language which I enjoy using (and building). This language has (1) await/async semantics to perform coordination between multiple people, (2) integrated queries for simpler thinking about state, (3) privacy as a first class citizen, (4) access control is baked in, (4) reactive formulas which lazily compute values, (5) simple message passing between user and the document, (6) modern features which make doing bad things impossible, (7) state can be rewound with a rewind, (8) no need to serialize state or think about networking, and (9) more

I’ve got my work cut out for me, but I’m starting the effort to find early adopters and build community that want to play as I push forward on my strategy since many of my investments are driven by pathologies.

I would deeply appreciate any feedback, comments, and questions here, via my new shiny email: jeff@adama-platform.com, or my discord channel