Austin (@Snackk_sol) header
Austin (@Snackk_sol) profile picture

Building in Public Everyday

Joined May 2021
Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)Photo by Austin (@Snackk_sol)
sub agents being released into my codebase
411
1,976
231
30,587
1,453,381
i don't want to ever work with people who don't use ai. why would i work with NORMIES i want to work with WIZARDS SPELL CASTING DIGITAL REALM BENDING WIZARDS YOU EXPECT ME TO BE SLOWED DOWN BY A MUGGLE ???? DON'T HATE ME BECAUSE YOU CAN'T CAST SPELLS YOU ARE NOT ON MY REALM
19
6
1
72
2,638
i love asking questions man im a curious mfer for real thats why i love LLMs so much i can just ask shit and get an instant answer and then it triggers more questions which get more instant answers and then i go into a deep rabbit hole of curiosity and now i'm an engineer
4
2
1
31
1,297
Austin retweeted
“Mom how did we get so rich?” “Your dad had the Claude max plan”
103
363
49
6,597
330,736
i miss the old opus one-shot the code opus call the right tools opus go straight to prod opus
50
93
11
2,527
89,156
Austin retweeted
change the name all you want, we’re just going to call it clawdbot
73
22
4
390
39,799
“ai will replace saas” “people will prompt their own apps” BRO MOST PEOPLE DON’T EVEN WANT TO TYPE/SPEAK INTO THE CHAT BOX dawg you are the MINORITY
73
53
20
1,664
119,128
Even if AI is one big bubble and it never improves from here, it would still be the greatest economic opportunity in history and have changed the world forever The amount of power Claude Code has unlocked for the average Joe is indescribable
105
30
5
586
23,520
Claude: "I've defined phase one of your project to be completed during weeks 1-4 and updated the spec." Me: "Implement phase one now." *5 minutes pass* Claude: "Phase 1 is complete."
172
152
39
4,108
489,312
1. Download @cursor_ai 2. Open project / make a folder 3. Maximize the chat 4. Remove all mental filters and just yap to it about any idea you you have, usually whatever you're making doesn't get interesting until like 20-30 minutes in of constant feedback 5. Just get reps
206
114
28
2,467
478,886
Austin retweeted
pro tip to vibecode UIs that do not look AI generated if you are NOT a designer search online for any "UI library" - look at examples, find the one you like the most find the main docs page, copy it entirely and create a file called [library_name].md instruct AI to use that library to build the UI and attach the file to the prompt - if the library is complex and there are multiple docs pages tell AI to crawl the documentation starting from the doc you gave it this improves your UI designs with virtually zero effort
32
54
2
1,342
78,395
Like, RT and follow to open this box 👀 Will pick out a few winners for a special reward 💜
265
370
6
669
88,658
2 months left until 2026 Most important 2 months of your entire life. If you use these months correctly I promise you'll dominate 2026 Here's the habits I'd implement right now: 1. Download the ChatGPT desktop app. Have it open 24/7. Weave it into everything 2. Download Claude Code if you haven't yet. Build your first app 3. Check out Google AI Studio's new vibe code mode. It's revolutionary. 4. Start filming videos about your passions. When AI eats all content, authentic human video is all that will remain (I've grown my YT to almost 50k the last year) 5. Make it your goal to launch 1 app by January 1st 6. Eliminate doom scrolling for the next 2 months. It is the single bad habit that will hold you back from greatness 7. No more casual drinking and drugs Do these 7 and I promise your friends will be shocked at who you transform into
70
89
3
1,209
104,406
I don't know who needs to hear this but if you have 15 extra minutes tonight, you can download Claude Code and by this time next week have a money generating app 0 programming experience required Find a way to not doomscroll for 15 minutes and you can legit change your life
107
39
5
698
73,550
Genuinely shocked how few people are clocking & understanding what’s happening in programming right now now 200 comments arguing “nooo you gotta understand every line of code” What do the words automation and delegating mean??? You don’t hire someone to clean your house then micromanage what brand cleaning spray they use, or watch exactly how they scrub a toilet. You just say “clean that room, be sure to get the counters”. The whole point is to hand it off, and work at a higher abstraction level. The AI can worry about the details. Rich people don’t grocery shop. Kim kardashian does NOT know how much a gallon of milk costs. Your goal with coding Ai agents is to not even think about the proverbial cost of milk in your codebase!! It will just happen because you told the AI to go grocery shopping for you. Then suddenly a home cooked codebase will appear. Poof.
Maintaining vibe coded apps is a job for GPT-7 not a human. Tech debt inflates away with every better coding model. The notion that you need to personally understand the code is hilariously short sighted, and clearly from someone who has never worked a large co with turnover.
353
80
48
1,524
331,854
Software Engineers are not paid for writing code. They’re paid for solving problems. The faster you accept this, the better your life and career will be.
461
1,608
154
17,346
855,828
The new steam age. This is actually becoming true in many cases. It's possible to do so much more on your own now.
542
4,200
347
41,356
2,020,185
Hearing from a lot of good founders that AI tools are writing most of their code now. Software engineers orchestrate the AI. They are also finding it extremely hard to hire because most experienced engineers have their heads in the sand and refuse to learn the latest tools.
497
153
144
2,877
893,179
One of my engineers just DM'd me his exact process for one-shotting very large codebase changes. I asked him if I could share it with you all. He said "send it." Quick disclaimer: 99% of his time is spent iterating on a markdown “plan” file and 1% is spent implementing. Here's his process... Step 1 - Building the Plan Prompt: • First I write a long essay in English that’s a stream of consciousness about all the technical and business requirements for the feature we’re building, concerns I have about implementation, etc. • Gather context either through RepoPrompt or dragging many files/folders into claude code • Often I include a “codebase map” file that maps out where everything is located • Paste in my “Code Plan” super prompt Step 2 - Generating the Plan: • The agent creates a new plan file broken down into discrete units of work • Each step references the full file paths and whether we’ll be creating, modifying, or deleting each file • A justification section for why this change is necessary and a short english explanation of the change • The full code snippet with every line that will be added and deleted • These plans can be 2000+ lines Step 3 - Manual Plan Review: • Manually review the plan and write a list of questions and comments, go back and forth with the agent to modify any major areas of concern Step 4 - Agentic Plan Review: • Paste in my “Plan Reviewer” prompt and the plan file and ask Claude to review the plan • Claude reads and analyzes every file dependency and outputs a list of blocking issues, suggested improvements, a rubric for the plan across 5 categories (i.e. “backwards compatibility), and a final grade for the plan out of 100 Step 5 - Plan Refinement: • After discussing the critical issues and suggested improvements to the plan, I paste in my “Plan Refinement” prompt that surgically edits the plan without adding bloat • I pass the updated plan through the review stage and repeat the process until the plan is at 90+/100 and Claude remarks about how awesome the plan is Step 6 - Plan Execution: • I pass the plan to a new Claude instance, paste in my “Plan Execution” prompt, and let it go on Auto-Accept mode. The agent tries to parallelize as much work as possible. • This is a good time for a bathroom break. Step 7 - Testing: • Run manual and automated tests • Most of the time, if I did the planning session well, everything works on the first try Step 8 - Documentation: • If we changed something major or added entirely new systems, I’ll have Claude write new documentation or update the existing project documentation Bonus - Context Handoff: • I don’t trust Claude Code’s “compaction” feature. If Claude is every running below 10% context, I’ll paste in my “Agent Handoff” prompt to create the perfect context package to kick off a new agent P.S. if you want more badass AI engineering workflows, make sure to follow @dan_zakon.
53
34
8
703
91,207