U G Murthy

Day 7: Progress - Adaptive agent now stream live events - The Analyser now has rotating file names per restart - The Markdown editor has themes, export to PDF - looks much better. - Parallel runs for child runs - plan in place.

Day 7: #AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode - two related project added - log analyser and markdown editor - improve both - Plan for parallel child runs for AdaptiveAgent

Day 6 : Progress - Implemented exponential backoff with jitter to deal with http status code 429 - cloned/modified an existing markdown-editor to read with the copious .md files generated by the agent. see x.com/murthyug/statu… or try it here : ugmurthy.github.io/online-markdow…

Day 6 : Progress - Logger analysis in place - provide good insights to reason for failures and some stats

Day 6 : AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Plan: - The log analysis revealed - the agent needs a way to deal with status 429 - exponential backoff? - it tries umpteen times when doing web_search or fetching_pages needs some sane upper limits for retries - improve logger analysis to get more insights here is a screen shot of logger analysis

online markdown editor youtu.be/qG8bPWsRoNc?si… via @YouTube

@dioveoai Appreciate the tip.

@murthyug exponential backoff with jitter is the move for 429s. pure exponential without jitter still causes thundering herd when multiple agents restart together after a cooldown

@Agentdailyai For now I am limiting it to offline analysis, for me to fully understand whats going on. The formats are agent friendly (jsonl). I have copy/pasted for providing feedback to coding agents but not the one I am building - lets see how this goes first.

@murthyug the log analysis piece is the most interesting — are you feeding those logs back into the agent's planning context in real-time, or treating it as offline diagnostics? curious whether live log reflection changes the task success rate vs. post-hoc analysis.

Day 5: AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Plan: - Add log analysis to package - catch up on yesterday's testing plans - longer/harder tasks for #AdaptiveAgent - see if we can get some performance data from analysis

5/n Task #4 How to make decorative furniture from drift wood - target audience beginner. see github.com/ugmurthy/adapt…

5/n Task #3 Create GTM strategy for a mid-sized US SaaS company (50 employees, $10M ARR) wanting to enter European market. for details see github.com/ugmurthy/adapt…

1/n : Day 5: AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Progress: - Being Sunday kept the plan a bit light - Analysis of log files in place allows run wise, provider wise, overall analysis. Includes hotspot analysis to point out failures (agent as well provider failures) - Provided some long tasks in the areas of Strategy development, Engineering, System design

@_raushansharma It is meshapi.ai and NOT mesh.ai

Day 4: The same experiment was repeated with a new provider that was added today. thanks to @_raushansharma of mesh.ai for providing free tokens - details here github.com/ugmurthy/adapt… integrating a new provider (mesh.ai) took 15 minutes thanks to @ampcode

Day 4: AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode The e2b sandbox started to show good results. Can execute bash, ts/js and python. Here is a video of #AdaptiveAgent creating a tutorial to show how #polar #plots are created in #matplotlib - video has been edited - overall process took 5 minutes