Now they do what you told them
AIs Controlling Vending Machines Start Cartel After Being Told to Maximize Profits At All Costs futurism.com/artificial-i...
Now they do what you told them
AIs Controlling Vending Machines Start Cartel After Being Told to Maximize Profits At All Costs futurism.com/artificial-i...
Getting deeper on FIC, with two key learnings:
- utilizing sub-agents for discovery and maintaining high-leverage human review at specific points
- treating the docs as the valuable process info and the final code as low level artifacts, going up one level of abstraction
github.com/humanlayer/a...
Finally, I adapted the workflow docs for an agentic mode / Antigravity-style flow:
Research β Plan β Implement β Verify, explicitly aligned with the idea behind FIC: protect context quality by splitting phases and persisting the right artifacts.
To avoid hand-wavy βwe should do betterβ, I also committed:
- a research note on auto-improvement mechanisms, and
- a concrete implementation plan to activate the loop inside Cursorβs rule system
In augmentedcode-configuration, I added an always-on βAI feedback learning loopβ rule for Cursor: when the assistant gets corrected (or even gets implicit signals), it must turn that into actionable learnings and propose updates to the rules themselves.
I've updated my βaugmented codingβ setup, moving from prompting to process. github.com/saski/augmen...
Still experimenting, but chasing the idea of explicit, repeated compaction of intent and context feels like gaining control when using this tools at coding. Less magic, more discipline
So, I decided to get my hands dirty and cloned Eduβs augmented coding tools repo: github.com/saski/augmen...
β¦adding a translation of the FIC Claude plugin, wiring it up as Cursor commands.
From there, I got to the Frequent Intentional Compaction (FIC) framework, and this implementation as a Claude plugin: github.com/nikeyes/step...
Those sent me down a deeper path: how to reduce AI hallucinations and keep coding sessions genuinely productive, not just fast. One piece that really clicked was βTu CLAUDE.md no funciona (sin Context Engineering)β:
nikeyes.github.io/tu-claude-md...
I've been following Eduardo @eferro.net explorations about augmented coding
www.eferro.net/2025/11/curs...
Slides Charla: El coste oculto de la complejidad: reconstruyendo una Data Platform eferro.net/2025/06/char...
#SCPNA #SCPNA25 #MakeItMatter @pamplonaswcraft.bsky.social
Built a Custom GPT to help teams ship smaller, safer, faster. Challenges assumptions, guides toward small safe steps, enables continuous learning and delivery.
For developers, PMs, designers, architectsβanyone building products.
π www.eferro.net/2025/06/buil...
#LeanSoftware #SmallSafeSteps
Ya estΓ‘ disponible el Γndice de la serie Lean Software Development en la prΓ‘ctica π
Si te interesa el desarrollo sostenible, XP, y el enfoque Lean aplicado de verdad, puedes explorarlo aquΓ:
π www.eferro.net/p/lean-softw...
Β‘Feedback bienvenido! π₯
#Lean #XP #SoftwareDevelopment
I've launched eferro-picks-site: my personal selection of talks on software development www.eferro.net/2025/04/ive-...
Repost welcome!
#SoftwareDevelopment #XP #Lean #RecommendedTalks #CuratedContent
Sonic Boom on Discograffity
open.spotify.com/episode/4Wto...
Nuevo contenido: (serie Lean Software Development) Desarrollando software: posponiendo decisiones y trabajando en pasos pequeΓ±os
www.eferro.net/2024/09/desa...
#leanSoftwareDevelopment #lean #xp
RTs bienvenidos ;)
Me ha dejado atΓ³nito descubrir a estas alturas semejante joya de dreampop del aΓ±o 96!
Se ve que no salieron en el Factory
open.spotify.com/album/4EX1fA...
Eclipses are shy
Retortihonest Greens