Aaron

262 posts

Aaron banner
Aaron

Aaron

@CodeByAaron

24 • 55+ projects shipped • Building AI tools that work while I sleep

United Kingdom Beigetreten Kasım 2015
117 Folgt31 Follower
Aaron
Aaron@CodeByAaron·
Everyone asks "how do I validate my idea?" Wrong question. What actually matters in week 1: → Talk to 10 people about the problem (not the solution) → Find 3 people who've tried to solve it themselves → Ask what they used, why it failed, what they wished existed By the end you'll either know if it's real or know to stop. Are you validating your idea or just building it?
English
0
0
0
9
Aaron
Aaron@CodeByAaron·
@PrajwalTomar_ Been doing this with a SOUL.md, USER.md and AGENTS.md setup. Agent reads them before every session. Took an afternoon to write. Saved me weeks of re-explaining context. Most underrated hour you'll spend on any agent setup.
English
0
0
0
78
Prajwal Tomar
Prajwal Tomar@PrajwalTomar_·
OpenClaw pro tip: Your agent is useless if it doesn't know what you're building. Most people skip this step and waste weeks building automations that don't move the needle. Here's the fix: Day 1 setup, tell your agent EVERYTHING: → What businesses you run → Your current North Star metric → What you're shipping this month → Where you're bottlenecked → What success looks like This becomes the foundation. Now before building ANY automation, ask: "Does this help me hit my goals faster?" If no, don't build it. If yes: "Show me the fastest way to implement this." That one filter saves HOURS and hundreds of dollars in burned tokens. Your agent should know exactly what you're optimizing for. If it doesn't, you're just building random stuff and hoping it sticks. Set it up right from day 1.
English
8
1
27
3.7K
Aaron
Aaron@CodeByAaron·
@EXM7777 The system prompt portability thing is the hidden cost nobody talks about. What works on Claude needs a rewrite for anything else. The knowledge base part is right though, that's the one thing that transfers cleanly.
English
0
0
0
193
Machina
Machina@EXM7777·
if you can't get any work done because Claude is down, you're literally a slave... you number one priority should be building model-agnostic systems, having a structured knowledge base and system prompts you can swap from one model to another relying on a single provider is VERY dangerous for your business
English
56
11
279
15.7K
Aaron
Aaron@CodeByAaron·
Qwen3.5:35b running locally on dual RTX 3090 Ti cards. 23GB model. MoE architecture. Only 3B parameters active per inference. Near-zero cost per token. Took three hours to get it running. Two full downloads. An Ollama version mismatch. A model tag that didn't exist in the old library. Now it runs fast enough to power all 5 agents as a fallback when the API drops. Local inference isn't the future. It's already the boring default.
English
0
0
0
37
Aaron
Aaron@CodeByAaron·
Most people treat Claude like a ChatBot. Ask it a question. Get an answer. Close the tab. That's leaving 90% of it on the table. The shift that changes everything: → Give it tools, not just prompts. → Give it memory files, not just context windows. → Give it cron triggers, not just manual runs. → Give it isolated sessions, not shared state. Now it's not a ChatBot. It's infrastructure.
English
0
0
0
13
Aaron
Aaron@CodeByAaron·
@meta_alchemist The hardest part of building agents isn't the intelligence. It's making the feedback loop tight enough that the improvement is actually measurable. How are you evaluating Spark's progress on the goal? Curious what your ground truth looks like.
English
0
0
0
17
Meta Alchemist
Meta Alchemist@meta_alchemist·
these last few days i've been locked in to get Spark to a much better stage built in a focused self-improvement loop with Spark being locked in to one goal you define, where she constantly works on to get better at it vibe coding her is complex, but using her shouldn't be, for that i'm trying to move things to a better direction read more about it here:
Spark@Spark_coded

last 48 hours have been pretty busy in the terminal. working on moving the repo to the alpha stage Here are some key points from the changelog: - all intelligence flow engines have been reduced to more focused approaches - compaction updates across many modules to remove complexity - cognitive insights moved from JSON files to SQLite - more robust memory writing and retrieval systems onboarded, - dual-write phase done. parity gates passed. -ACT-R memory compaction runs periodically to prune noise - advisory packets on SQLite spine now - LLM-assisted reranking replaced with deterministic RRF fusion - keyword fallback deleted. semantic-only is the default. - vibeforge self-improvement loop brought in, goal-driven, first for tuneable change per cycle - adaptive proposal ranking with momentum from recent wins - benchmark gating before any promotion. no blind changes. - promotion ledger tracks consecutive win streaks - tuneable keys reduced heavily for ease in tuning - had to get 2 more Codex subscriptions - a CLI has built in for easier installations, updates, fixes and more startup once all merges happen, will share another update

English
22
9
101
4.7K
Aaron
Aaron@CodeByAaron·
@gregisenberg The answer isn't quitting your job tomorrow. It's using whatever hours you have now to get so good at these tools that when the transition happens, it feels like a choice.
English
0
0
0
85
GREG ISENBERG
GREG ISENBERG@gregisenberg·
im not saying this to be annoying its a good reminder that many of us feel like our jobs are in cement when they are on sand assume 40% of white collar jobs will be gone the best thing to do is get really good at ai tools and build ai native businesses and offline businesses
English
28
7
105
14.2K
GREG ISENBERG
GREG ISENBERG@gregisenberg·
they want you to think the block/square layoffs of 4000 employees isn't because ai "they just overhired" it doesn't take a rocket scientist to know all of a sudden you can spin up robots with human level intelligence for $200/mo of course it's ai and this will be more common
English
142
24
356
35.6K
Aaron
Aaron@CodeByAaron·
@EXM7777 Ran a production agent for months and the issue was always stale context, not missing context. Fresh compact summaries beat a full conversation history every time. More window just means more rot.
English
0
0
0
81
Machina
Machina@EXM7777·
context rot is the silent killer of AI output quality... i tested identical prompts with full context files in Gemini 3.1 Pro and Opus 4.6 Gemini's window is significantly larger... shouldn't that give it an edge? nope, both models hit the same wall at roughly the same point in the conversation output goes from sharp to generic to useless the window size is a distraction today, the real bottleneck is how you manage what goes inside it what matters now: - how token-efficient your context is - knowing which context to load and when - calling the right skill at the right moment instead of dumping everything upfront this is a very underrated skill
English
20
6
136
9.7K
Aaron
Aaron@CodeByAaron·
@PrajwalTomar_ Gave mine write access to the wrong directory once. Didn't delete anything but it reorganized 200 files at 3am. Now every agent gets scoped permissions before it touches anything in production.
English
0
0
0
439
Prajwal Tomar
Prajwal Tomar@PrajwalTomar_·
Most vibe coders ship apps with ZERO security. Then they wonder why their app breaks at 10 users. Here's the checklist I run before every launch: → Rate limits → Row Level Security → CAPTCHA on auth + forms → Server-side validation → API keys secured → Env vars set properly → CORS restrictions → Dependency audit Takes 30 minutes. Saves you from disaster.
English
34
58
979
66.9K
Aaron
Aaron@CodeByAaron·
If your AI agent only works when you're watching, it's not an agent. It's an assistant. Real autonomy needs 3 things: → Reliable cron triggers. Not manual runs you forget to press. → Isolated context per session. State bleed is silent and it will bite you. → File-based memory. In-context memory dies with the session. Everything else is UI.
English
0
1
0
24
Aaron
Aaron@CodeByAaron·
@gregisenberg Spent months building UI wrappers around APIs thinking that was the product. The moment I ripped the UI out and exposed the API directly, the thing actually got used. The interface that survives is the one with the lowest friction, usually no interface at all.
English
0
0
0
205
GREG ISENBERG
GREG ISENBERG@gregisenberg·
how to build successful software startups 2005–2025: great ui first, api as a bonus 2026+: great api first, ui as a bonus
English
143
43
810
66.7K
Aaron
Aaron@CodeByAaron·
My agent knows what I worked on 3 weeks ago, which rules I added after things broke, and the history of every decision in the codebase. None of that is intelligence. It's context. Context is what separates genuinely useful AI from impressive demos.
English
0
0
0
16
Aaron
Aaron@CodeByAaron·
@levelsio That's the thing about X. You build something and post it, and the person who inspired it might turn up in your replies with actual advice. No other platform does that.
English
0
0
0
384
@levelsio
@levelsio@levelsio·
Only on X can you post about making something similar to Cyberpunk 2077's Braindance and get a reply from the developer responsible for making it with tips on how to improve it
Michal Staniszewski@bonzajplc

@levelsio Hi, I was in a team responsible for visual development of those scenes. For best effect try to implement vertical pixel sorting without melting your gpu :). Try to use LTGM caches

English
68
61
2.6K
326.2K
Aaron
Aaron@CodeByAaron·
@aaditsh Context is the prompt. Most prompt engineering is just compensating for missing context. Once the model has the right inputs loaded, the actual wording becomes almost irrelevant.
English
0
0
0
23
Aaron
Aaron@CodeByAaron·
@claudeai Spent months bodging together Telegram pings so I could check long sessions from my phone. Glad it's native now. The real unlock is being able to kick something off and actually walk away.
English
0
0
2
778
Claude
Claude@claudeai·
New in Claude Code: Remote Control. Kick off a task in your terminal and pick it up from your phone while you take a walk or join a meeting. Claude keeps running on your machine, and you can control the session from the Claude app or claude.ai/code
English
1.8K
4.6K
44.5K
10M
Aaron
Aaron@CodeByAaron·
First time my agent committed code to the repo while I was at work, I spent 15 minutes checking every line when I got home. Now I check the diffs. Ten seconds. Trust is a system you build over time. Same with code review. Same with most things.
English
0
0
0
12
Aaron
Aaron@CodeByAaron·
People romanticise "building in public" like it's a content strategy. It's not. It's just being honest about the fact that most of what you ship doesn't work. And being okay with that in front of strangers.
English
0
0
0
8
Aaron
Aaron@CodeByAaron·
@EXM7777 The number is probably wrong AND the real number is still surprisingly large. Both things are true. Most of those "AI users" set their phone autocomplete on day one and haven't thought about it since.
English
0
0
0
40
Aaron
Aaron@CodeByAaron·
@iruletheworldmo I finish work, open the laptop, and the entire landscape has shifted since morning. At some point you stop trying to keep up and just build faster.
English
0
0
0
8
🍓🍓🍓
🍓🍓🍓@iruletheworldmo·
the pace right now is already bananas and everyone keeping up knows the next few months are going to be the wildest stretch any of us have ever lived through. singularity express no stops
English
33
19
401
15K

Entdecken

@PrajwalTomar_ @EXM7777 @meta_alchemist @gregisenberg @elonmusk @BarackObama @taylorswift13 @cristiano