Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
85 contributions to Clief Notes
⚙️ New in David's Corner. A two-part advanced piece: Tech Specs Rubric Creation.
It opens with a problem we all pretend we solved. You graded the final answer and it looked fine. Two steps upstream, the plan was already wrong. End-to-end testing hides that. Errors compound silently across a run, and the last node takes the blame. The spec builds something better: 🔹 Part 1. Model the run as a DAG. Type-specific rubrics for every step. Calibrated LLM judges, cross-family only, so a model never grades its own kin. 🔹 Part 2. A 3-level failure taxonomy, root cause attribution across the graph, and CI gates that block regressions before they ship. The numbers are concrete. 63.2% of step failures are propagated from upstream. 2.17x the failure detection of end-to-end grading. Root cause triage from 4.2 hours to 22 minutes. Advanced reading, not a tutorial. Part 1 links to part 2. https://www.skool.com/cliefnotes/classroom/c7f102c7?md=5a4fa9d585e541058903250ecdd2e367 https://www.skool.com/cliefnotes/classroom/c7f102c7?md=51b7ef5e52394768b43f63eaee2a0896 Your move: if you run multi-step agents, how would you know today if step 2 of 5 was the broken one? Tell me below. The comments are the course. The text is fuel.
⚙️ New in David's Corner. A two-part advanced piece: Tech Specs Rubric Creation.
2 likes • 1d
Thanks @David Vogel definitely saving your share!!
SGM, the method behind the archive I work in every day
Hi everyone, In this repository I share part of my work: the method behind the archive I use every day with AI agents. It decides what an assistant is allowed to remember about my work, and who answers for it when what it remembered turns out to be wrong. https://github.com/Valmor-R-Reichel/SGM If you want the why, this page tells how it was built and the criterion behind each step. The scripts and tests were written by AI agents at my request, so the page is about what I decided to measure and what I count as an error. https://github.com/Valmor-R-Reichel/SGM/blob/main/docs/criterion-over-repertoire.md If you like the philosophy behind it, the two essays I shared here are that side of the work. https://www.skool.com/cliefnotes/the-social-cost-of-asking-another-person https://www.skool.com/cliefnotes/when-the-flashlight-looks-like-the-sun The repository moves slower than the archive. I work on it when I can, because each piece has to be translated into only the method, without the people and the data from my job. The README still describes the system as it stood in September, and the changelog lists what changed since. The latest entry is from yesterday. I rewrote the main file my assistant reads in every conversation and counted my corrections before and after. The number did not visibly move, and the first judge almost reported an improvement that was not there. https://github.com/Valmor-R-Reichel/SGM/blob/main/CHANGELOG.md
1 like • 1d
Thanks for the GitHub shares!!!
Retrofitting a complex, long-term Go IAM project to ICM (and stopping AI loops)
I’ve been building VK Auth (an idiomatic Go Identity & Access Management engine) for around a decade. Recently, I retrofitted the entire codebase onto our ibe-workflow-template (first introduced on this forum) — combining Jake’s ICM principles, Karpathy’s Claude rules, and Vigil Keep’s Intent-Based Engineering (IBE). It wasn’t smooth sailing out of the gate, and solving the biggest failure mode taught me a lesson worth sharing here. 🏊‍♂️ The Failure Mode: Swimming Halfway Across the River When migrating core architectures, the AI kept suffering from context confusion and mechanic fixation. The hardest problem was moving from single atomic writes on the authorization tree to a phased write mechanism that never violates invariants (yielding atomic writes in the best case and eventual consistency in the worst). Because this mechanism has irreducible complexity—the entire structure has to be in place before any of it works—the AI would get halfway through the refactor, lose its footing when intermediate checks failed, get disoriented, and swim right back to the shore it started from by reverting to the old architecture. 🛠️ The Fix: A 3-Way Resolution Mechanism (No Looping) To stop the agent from thrashing or reverting at the implementation layer, I enforced a strict three-step resolution ladder with zero looping: 1. Forward Drive: Intent → Decisions (ADRs) → Design / Specs → Implementation 2. Backward Validation: Implementation is validated directly against Intent. When something breaks, feedback walks backwards up the ladder to the first rung capable of answering it: - Can't solve it in Implementation? There is a gap in Design / Specs. - Can't solve it in Design / Specs? There is a gap in Decisions (e.g., missing trade-off priority between performance and consistency). - Can't solve it in Decisions? You have conflicting or non-resolvable Intents. 3. Forward Remediation: Once the gap is patched at the right rung, execution drives back down to Implementation and concludes. If it still fails, it stops and mandates human intervention—no endless agent loops burning context.
1 like • 1d
Thanks for the share.
The Difference Between AI Slop and AI That's Actually Useful
AI slop I've seen: Generic AI-generated sales emails. We've all seen the ones that start with something like "I hope this email finds you well" and go on for three paragraphs without saying anything meaningful. What felt off: They're polished, but impersonal. You can tell someone generated a message and sent it without really understanding who was receiving it or why they should care. AI being used well: I've been building two AI agents for BashTime, a higher-ed software company I'm involved with. Rex researches and qualifies potential customers, while Piper helps draft personalized outreach. We recently changed our approach to focus on specific challenges educators face rather than sending generic product pitches. Our first batch of five emails generated two responses within three hours. What made it different: AI did much of the research and drafting, but we reviewed the prospects, refined the messaging, and made the final decisions ourselves. The emails were aimed at starting relevant conversations, not just getting something into someone's inbox. My takeaway: AI makes it easy to produce more content. That doesn't mean the content is worth producing. The real value comes from combining AI's speed with human experience, judgment, and a willingness to revise what's not working. I'd rather send five thoughtful emails than fifty generic ones.
0 likes • 1d
Thanks for the share.
From AI Prompts to Persistent Workflows
How I currently use AI: I've moved well beyond one-shot prompts. I use Claude Code, ChatGPT, and other tools to manage projects, build AI agents, and maintain a GitHub-based Second Brain. My AI Workspace uses structured Markdown files to preserve project context, decisions, instructions, and current status between sessions. The limitation I've noticed: Even with this structure, my AI assistants don't always retrieve the right information or consistently follow established workflows. My personal AI assistant, Donna, is a good example. I've spent considerable time improving how she prepares for and conducts our daily scrum. What I want to explore: The lesson's discussion of memory and retrieval, particularly the difference between simple file-based context and approaches like knowledge graphs. Why: I'd like my AI assistants to find relevant information across projects and sessions without needing everything loaded upfront. But I also don't want to add another layer of technology unless it solves a real problem. My first step: I'll identify one recurring situation where Donna struggles to find or use the right context. Then I'll test whether better file organization or retrieval instructions solve it before considering a more sophisticated memory system.
0 likes • 1d
Thank you for sharing your experience. It help knowing the types of challenges people are experiencing.
1-10 of 85
Terrence Rideau
4
29 points to level up
@terrence-rideau-2028
I am an AI & GenAI Learner!!!

Active 10h ago
Joined Apr 25, 2026
Powered by