Detected Across Public Sources
Evidence specific to Tabnine across indexed public receipts.
X
SOP driven problem solving for using LLM tools like @windsurf_ai @cursor_ai @tabnine - based on 350+ hours of playing with them over the last few months - Create a new folder for context retention docs called aa_ref_01 - Create file named r01_initial_ideation.txt for just noting down all the ideas without any structure - Create file named r02_domain_analysis.txt which will contain the best of the conversation we have with tab nine or @web in order summarize the best domain knowledge needed for the problem statement - Create file named r03_prd_v1.txt which will be the first PRD for the user journey at a high level - Create file named r04_prd_v2.txt which hopefully is a more precise version of the previous PRD based on multiple iterations on the previous file - Create file named r05_arch_L1.txt which will contain the first attempt at the architecture of the PRD - this file should leverage a relationship of consistency with r04_prd_v2.txt + r02_domain_analysis.txt PAUSE HERE & NOTICE: a. LLMs should be treated as search + assimilation tools, you have to bias them with right keywords b. LLMs need to iterate to arrive at a more refined output c. LLMs need to sumarize information which will lost by the sliding context window, hence you constantly create new docs which are in sync with previous docs now - Create file named r06_arch_L2.txt which will contain the second iteration of the architecture of the PRD - this file should leverage a relationship of consistency with r05_arch_L1.txt + r04_prd_v2.txt + r02_domain_analysis.txt - Create file named r07_explain_architecture.txt which will contain the explanation of the architecture of the PRD - this file should leverage a relationship of consistency with r06_arch_L2.txt + r04_prd_v2.txt + r02_domain_analysis.txt. This file is basically a rubber-duck debugging of the architecture with us as the listener who wants to make sure that the r06_arch_L2.txt is well thought out - Create a file named r08_TDD_v1.txt which will basically ask the LLM to write a TDD plan to implement the r06_arch_L2.txt in the most minimalistic way possible, it should be consistent with r04_prd_v2.txt & r02_domain_analysis.txt - If the r08_TDD_v1.txt is good enough, then select each step of the TDD plan as a checkbox - and ask the strong LLM (code LLM) to implement the TDD plan in the most idiomatic bug-free way possible line by line, thoroughly following the TDD methodology of starting with Red Test, then Green Test, then Refactor - Create a file named r09_bug_log.txt to note copy all the bugs encountered in the process of implementing the TDD plan - Create a file named r10_avoid_bugs.txt which will contain meta patterns and best practices discovered during the implementation of the TDD plan - Create UAT plan for the whole feature, ideally this should be very close to r04_prd_v2.txt - and manually check the results, and document them in r11_uat_results.txt - Create a file named r99_unclassified.txt to note down all the prompts, ideas, mistakes, lack of knowledge or references or links which you encounter during the journey with timestamps PAUSE HERE & NOTICE: d. LLMs need to be asked to reason themselves so they can find their own logical fallacies, like rubber duck debugging e. LLMs need to be given reference files for patterns to avoid, be it via past bugs or anti patterns to avoid f. LLMs need to follow TDD because it makes it easier to arrive at the final code - it is an easier prediction problem if you know what will be the requisite final pattern, also helps you avoid mistakes early enough g. LLM whispering is a cumulative skill so constantly document your prompts, steps & learnings p.s. most won't be able to follow the above steps and would rather smash "fix this" every 2 minutes because they lack the patience to follow these steps methodically@amuldotexe · 74 engagement
A wave is coming. And nothing will stop it. Artificial intelligence no longer just answers questions. Now it writes code. And more and more, it executes it. The giants are already on the battlefield: @OpenAI, @github, @AnthropicAI, @CursorAI, @Tabnine — Generate code, but don’t execute or host it. @Google, @awscloud, @replit, @cognition_labs Generate, execute, and host the code — but within centralized cages. @dfinity with #Caffeine, built on #InternetComputer — Generates, executes, and hosts the code entirely on-chain. No servers. No permissions. No owners. This wave is inevitable. But only one path leads to true technological sovereignty. The rest… are just more efficient versions of the same prison. On June 3rd, for the first time, intelligence will run free. To the entire #ICPCommunity and our global reserve army: It’s time to leave the trenches and ready the memes. The countdown to #CaffeineDay has begun. #InternetComputer #SovereignAI #SelfWritingInternet #CaffeineAI #DDay #InDomWeTrust@iphelipevr · 69 engagementArticles
Commonly Appears With
Collapsed relationship overview for tools appearing with Tabnine.
No related-tool graph receipts are indexed for Tabnine yet.
Workflows
Three-plus-tool stacks involving Tabnine.
No three-plus-tool workflow receipts are indexed for Tabnine yet.
Micro Workflows
Exact two-tool connections involving Tabnine.
No exact two-tool micro workflow receipts are indexed for Tabnine yet.