"Modern LLMs exhibit functional Theory of Mind—specifically epistemic state tracking and second-order perspective-taking—which allows an external auditor agent to reconstruct another agent's false beliefs from a raw transcript without sharing its tunnel vision." Discuss.
Randal Schwartz
@randalschwartz.bsky.social
Google Developer Expert in Dart/Flutter & Project Lead for BlocSignal ecosystem. Five decade industry veteran.
An LLM memorized 10T tokens of static code (What) yet lacks Why and How. How aphantasia, 50 years of craft, and a Two-Track "magical voice" bridge the gap: dev.to/gde/the-pris...
The Prisoner in the Library: Mary's Room, Aphantasic Code in Motion, and the Two-Track 'Magical Voice'
Why does an LLM that memorized 10 trillion tokens of books still act like a reckless novice in your codebase? Mary's Room, Why/How/What vs. Be/Do/Have, aphantasic kinesthetic flashbacks, and the Two-T...
dev.to
🔴 #HumpdayQandA and Live Coding at 5pm BST / 6pm CEST / 9am PDT today! Answering your #Flutter and #Dart questions with @devangels.london @randalschwartz.bsky.social and @johnwiese.bsky.social www.youtube.com/watch?v=FhKL... #FlutterCommunity #Flutter #Dart
YouTube video
youtube.com
Why do AI coding agents still write .withOpacity() and fake widget test mocks in Flutter? In my Flutter Community AI Circle keynote, I walk through how 626 Synthetic Scars hit 0.0% repeat regressions across 266 tickets: www.youtube.com/watch?v=SSM5... #Flutter #Dart
FCAIC #30 - Autonomous Flutter Agents: Senior Engineer Reflexes
YouTube video by Flutter Community
youtube.com
Stop asking LLMs to do calculator math in GenUI. Client functions in bloc_signals_genui compile into computed() signals updating in 0ms at 120 FPS. Deep-dive: dev.to/gde/stop-ask... #Flutter #Dart
Stop Asking LLMs to Do Math: Client-Side Functions in Flutter GenUI with bloc_signals
Why Andrew 'Red' Brogdon's insight on client-side functions is critical for Generative UI—and how bloc_signals_genui turns catalog functions into 0ms synchronous reactive signals.
dev.to
Why does "vibe coding" fail in production? Because it demands "Have" (Content) without "Be" (Context). Part 4.4 of Synthetic Scars—ladder rungs, the "Null-State" couch flash & stand-up comedy: dev.to/gde/being-do...
Being, Doing, Having: Why 50 Years of Senior Craft Lives in Context, Not Code
Why does standard AI code generation try to jump straight to 'Have' without the proper 'Be'? How ladder rungs, the 'Null-State' couch flash, and stand-up comedy explain what separates senior engineers...
dev.to
Why can a crow bend wire into a hook, while an LLM flies straight into the bug zapper? Part 4.3 of Synthetic Scars explores the Genomic Bottleneck & AI instinct: dev.to/gde/crows-mo...
Crows, Moths, and Makefiles: How 300 Megabytes of Crow DNA Explains Why a 20-Kilobyte Scar File Outsmarts a Terabyte LLM
Why do New Caledonian crows bend wire into hooks with zero training while a trillion-parameter LLM flies straight into the same production bug zapper? How Anthony Zador's Genomic Bottleneck explains w...
dev.to
🔴 #HumpdayQandA and Live Coding at 5pm BST / 6pm CEST / 9am PDT today! Answering your #Flutter and #Dart questions with @devangels.london @randalschwartz.bsky.social @johnwiese.bsky.social and @makerinator.bsky.social www.youtube.com/watch?v=92Mb...
YouTube video
youtube.com
Why did 6 AI prompt grammars collapse into one 1980 Smalltalk-80 keyword cascade? In Suite 8 (360 trials), Prose hit 2.5%, Lisp hit 91.7%, and Smalltalk-80 hit 100.0% (120/120)—plus 7 live production runs (0/250 repeat regressions). dev.to/gde/the-post...
The Postcard That Ate Six Grammars: 1980 Smalltalk-80, 1976 Make, TDD, and the Four Attractors of Agentic Coding
Why maintaining six separate parsers for Make, Gherkin, EBNF, DOT, SysV, and prompt checklists collapsed into a single Smalltalk-80 keyword cascade—and how 40-year-old Xerox PARC technology gave AI co...
dev.to
Junior devs trust all 10 links on Page 1 of Google; a 40-yr senior crosses 4 off on sight (-∞). Yet LLM weights ARE popularity! Why AI labs can only bake in half our 600 scars: dev.to/gde/all-that...
All That Glisters Is Softmax: Shiny-Object Interns, Airplane-Magazine Execs, and Why AI Labs Can Only Bake In Half the Scars
Why do an eager intern with a Hacker News blog post, a C-level executive stepping off an airplane with an in-flight magazine, and a trillion-parameter LLM all walk straight into the exact same product...
dev.to
"My AI burns my 7-day quota in 3 days" & "My AI ignores my style guide" are the exact same attention bug: pasting // DON'T examples primes the KV-cache with the code you want to ban. Part 4 of Synthetic Scars—4-Vector Scars, 1991 Graphviz DOT, & the 6-Pillar Critic: dev.to/gde/giving-a...
Giving AI Pain: The Architecture of Synthetic Scars & The Rapid-Regret Miner
Why 90% of AI coding frustration comes from the Quota vs. Prompt-Compliance Death Loop, why Rich Sutton's Bitter Lesson fails in open-world software, and how we compile mid-flight collisions into 4-Ve...
dev.to
Most Clean Code is homage to a no-longer-necessary ritual of layers made obsolete by modern practices. Discuss.
Harvard studied 718 companies: AI agents write +30% more code and close 0 extra tickets because PR reviews jump +49%. I ran my Makefile workflow on Gemini, Claude & GPT-OSS: dev.to/gde/926-step...
926 Steps, Two Lobotomies, and Zero Nudges: Flight-Testing the Commitment Machine Across Gemini, Claude, and GPT-OSS
A 2026 Harvard study of 718 companies found that AI agents write 30% more code and close zero additional tickets because review times jump 49%. Here is what happened when I put my 1986 Commitment Mach...
dev.to
🔴 #HumpdayQandA and Live Coding at 5pm BST / 6pm CEST / 9am PDT today! Answering your #Flutter and #Dart questions with @devangels.london @randalschwartz.bsky.social www.youtube.com/watch?v=Kw2a... #FlutterCommunity #Flutter #Dart
YouTube video
youtube.com
Shipped bloc_signals_genui & bloc_signals_genui_flutter 1.0 on pub! ⚡ Built on Google's official A2UI (`a2ui_core`) protocol: • 0ms two-way WritableBinding forms & CheckRule validation • Pure Dart (Material 3 + ANSI TUIs!) • Riverpod & BLoC interop dev.to/gde/generati... #Flutter
Generative UI Without the Async Tax: Introducing bloc_signals_genui 1.0
Why we built a pure-Dart, 0ms synchronous Generative UI state machine on top of Google's official A2UI protocol—supporting two-way reactive forms, Riverpod/BLoC interop, Flutter Material 3, and ANSI t...
dev.to
Why do AI agent prompts turn into 130-line legal contracts—and still get ignored? Part 3.10 of Synthetic Scars: rewriting our agent kernel as a 1986 Flores Commitment Machine: dev.to/gde/requests...
Requests, Promises, Assertions, and Declarations: Rewriting Our AI Agent Workflow as a 1986 'Committed Speech' Machine
Why do 200-line agent prompts, Asimov's Three Laws, and sequential pre-landing checklists all fail the same way? How Fernando Flores's 1986 Committed Speech, Nuel Belnap's 1977 4-Valued Logic, and Ton...
dev.to
Why do LLMs invent APIs that never existed? Agents hallucinated because they were committed to the prompt. How Fernando Flores's 1986 Committed Speech and my 1999 spew BNF un-parser explain it—and the 1-line fix: dev.to/gde/if-you-n...
If You Never Do Anything Else to Fix Your Prompts, Do This One Thing
Why do LLMs confidently invent APIs that have never existed in the history of the universe? It is not a memory defect. Agents hallucinated because they were committed to the prompt—and Fernando Flores...
dev.to
Stop inventing prompt DSLs (Freq ≈ 0). Across 1,680 A/B trials, 50-year-old Unix/CS primitives (Makefile, /etc/init.d, EBNF, 2PC, Circuit Breakers, git bisect) crushed English rules: dev.to/gde/what-llm...
What LLMs 'Know' That You Don't Know They Know: Stop Inventing Prompt DSLs and Ride 50 Years of Unix Pre-Training Gravity
Why did a 1976 Makefile (Part 3.6) and a 1983 Unix init.d directory (Part 3.7) outperform 2,000 lines of English prompt rules? Because prompt engineering is static symbol linkage against 50 years of p...
dev.to
When an AI coding session hits 200k tokens, IDE auto-compaction summarizes 230k tokens of invariants into generic mush. How 1983 Unix /etc/init.d runlevels, fork()/wait(), and Memento tattoos made our agent compaction-proof: dev.to/gde/survivin...
Surviving the 200k-Token Lobotomy: How Unix init.d and 'Memento' Made My AI Coding Agent Immune to Context Compaction
Why host auto-compaction silently lobotomizes long-running AI coding sessions, and how 1983 SysV init.d runlevel files, Memento tattoos, and fork/wait subagents let our orchestrator survive two 230k-t...
dev.to
Why do 20-step AI prompts fail mid-PR? LLMs have no loop counter. Stuart Feldman solved this at Bell Labs in 1976: give your agent a Makefile DAG. dev.to/gde/stuart-f... #AI #Flutter #Coding
Stuart Feldman Was Right in 1976: Why Your AI Agent Needs a Makefile, Not a 20-Step Prompt
How replacing fragile linear prompt checklists with a 50-year-old declarative build graph eliminates agent amnesia, enables seamless mid-flight session resumption, and mechanizes software engineering ...
dev.to
Telling an AI what to fix triggers an "insecure neurosis death spiral." How 40 years of scars & 30 years of mentoring yield 95%+ first-pass architecture with Socratic nudges: dev.to/gde/nudging-...
Nudging with Questions: Why Telling Your AI What to Fix Triggers an Apology Death Spiral (And How to Advance Juniors)
What I learned mentoring juniors for four decades seems to apply to today's agentic coding: why barking code fixes triggers sycophantic panic, and how Socratic questions yield 95%+ first-pass success.
dev.to
BLoC discipline vs Signals speed? A Cubit or Bloc is simply a container that holds and protects a Signal. Keep events and 0ms computed reads; discard streams and microtask lag: dev.to/gde/a-cubit-... #Flutter #Dart
A Cubit or Bloc Is Just a Container Holding a Signal: How Much BLoC vs. Signals Do You Actually Need?
Demystifying BlocSignal through one irreducible mental model: keeping declarative intent and concurrency boundaries on the front, while letting reactive signal graphs power 0ms reads on the back.
dev.to
Exhausted barking line-by-line fixes at an AI agent? Synthetic Scars Part 3 explores the physics of Socratic prompting: non-Von Neumann ponds, simulated annealing, & chess Alpha-Beta: dev.to/gde/the-phys...
The Physics of Socratic Prompting: Somatic Recoil, Chess Alpha-Beta, & The NLP Meta-Model
Why scalar RLHF fails at credit assignment, why barking direct code fixes turns you into an exhausted human compiler, and how Socratic questioning plus topological scars prune doomed search trees at d...
dev.to
Why was Riverpod needed, and why don't you need it any more? In 2019, Flutter rejected PR #33213. InheritedWidget couldn't unlisten from conditional reads. The paper trail & the signals fix: dev.to/gde/the-sand...
The Sand in the Oyster: Why Riverpod Was Needed, and Why You Don't Need It Any More
The hidden mechanical flaw in Flutter's InheritedWidget that forced the creation of Riverpod, why escaping the widget tree created its own architectural taxes, and how push-pull signals finally resolv...
dev.to
🔴 #HumpdayQandA and Live Coding at 5pm BST / 6pm CEST / 9am PDT today! Answering your #Flutter and #Dart questions with @devangels.london @randalschwartz.bsky.social @flutterfly.bsky.social @johnwiese.bsky.social and @makerinator.bsky.social www.youtube.com/watch?v=jM_a...
Humpday Q&A/AMA with Live Coding :: 23rd September 2026 :: #HumpdayQandA #Flutter #FlutterCommunity
YouTube video by Flutter Community
youtube.com
Why do AI coding agents write flawless code at 2 PM on staging, but crash at 3 AM in production? Because software isn't continuous calculus—it's discrete hostile logic. Part 2 of Synthetic Scars is live on @devto: dev.to/gde/why-ai-c...
Why AI Coding Agents Crash at 3 AM: The Happy-Path Mirage & The Forced Continuity Defect
Why LLMs fail in production, why "more RLHF" cannot fix it, and how transferring 3 AM pager-duty trauma gives autonomous coding agents real survival instincts.
dev.to
🔴 #HumpdayQandA and Live Coding at 5pm BST / 6pm CEST / 9am PDT today! Answering your #Flutter and #Dart questions with @devangels.london @randalschwartz.bsky.social @flutterfly.bsky.social @johnwiese.bsky.social and @makerinator.bsky.social www.youtube.com/watch?v=m8Mq...
Humpday Q&A/AMA with Live Coding :: 16th September 2026 :: #HumpdayQandA #Flutter #FlutterCommunity
YouTube video by Flutter Community
youtube.com
Humans survive outages because our brains carry somatic scars. AI has no nervous system—every day is Groundhog Day. Here is how teaching AI pain gives it real wisdom: dev.to/gde/why-ai-k... #AI #SoftwareEngineering #Dart #Flutter
Why AI Keeps Making the Same Coding Mistakes—And How Teaching It Pain Gives It Wisdom
Frontier LLMs excel at textbook code but fail in production. Here is how Synthetic Scars and biological pain give autonomous coding agents real wisdom.
dev.to