Reflection on AIBussin — AI applications, systems and books

Critique, Revision, and Acceptance

Sat, 08 Aug 2026 16:23:00 +0100

Selection can only choose among the candidates it is given. A strong evaluator may recognize that every available candidate is poor, but it cannot select a correct answer that the generator never produced.

That limitation becomes practical because the candidates usually come from one model answering one prompt, and such samples correlate. When four candidates share the same misreading of the evidence, ranking them can still produce a confident winner while leaving the shared defect untouched.

The Thinking Agent

Mon, 01 Jan 0001 00:00:00 +0000

In the last chapter, you built a small agentic workflow in conversation. It had roles, visible state, a tool boundary, and a review loop.

Now we can strengthen the idea that made it useful.

generate
 ↓
critique
 ↓
revise
 ↓
compare
 ↓
accept, reject, or roll back

This is the architecture of a thinking agent: not because the system has private consciousness, and not because a model’s self-critique is automatically reliable, but because the system can externalize an attempt, inspect it, produce a candidate improvement, and decide whether the candidate deserves to replace the current version.

Companion Agents

Mon, 01 Jan 0001 00:00:00 +0000

Using Agents to Support a Life, Not Run It

Before we talk about companion agents, we need to talk about the practice that makes them useful.

The practice comes first:

a few words in the morning about what matters today;
a short evening reflection on what happened;
a note about energy, mood, focus, or friction;
a saved goal, routine, question, or decision.

You do not need AI for any of this. A notebook can do it. So can a text file, a calendar, Notion, Obsidian, or a private journal.

Let the Program Reflect

Fri, 28 Aug 2026 11:00:00 +0100

Chapter 12 ended on a specific limitation. MIPROv2 selected a candidate that deleted regularly from a regulated claim, and nothing in the pipeline could have told it otherwise, because the only signal it received was 0.967.

candidate
 ↓
metric
 ↓
0.967

A number can rank candidates. It cannot say which word was the problem.

The next mechanism replaces the scalar with language:

 flowchart LR
 P[prediction] --> M[metric]
 M --> SF[score + diagnostic feedback]
 SF --> RF[reflection]
 RF --> IM[instruction mutation]
 IM --> CA[candidate]
 CA -->|re-run| P

GEPA is DSPy’s evolutionary, reflection-driven optimizer. It reads per-predictor feedback, uses a separate reflection model to propose instruction edits, maintains a population of candidate programs, and returns the best one it found. The engineering idea underneath is smaller than the API:

Upgraded: The Beginning of Your Journey

Fri, 28 Aug 2026 10:00:00 +0100

“The most effective way to manage change is to create it”

You’ve seen what this process looks like.
You’ve seen what it can do.

You now have everything you need to begin.

Not a perfect method.
But a repeatable one.
Not a single answer.
But a way of thinking that brings out your best and makes the AI better, too.

 flowchart TD
 Start([📚 You Have the Tools]) --> A[🧠 Set Intention Clearly]
 A --> B[🔄 Move Through Structured Loops]
 B --> C[🔁 Adjust with Each Pass]
 C --> D[⚡ Recognize & Amplify Breakthroughs]
 D --> E[🎯 Pick Something Real]
 E --> F[🗣️ Start Talking / Prompting / Shaping]
 F --> G[🔁 Loop & Adjust]
 G --> H{Amplified Creation?}
 H -- Yes --> I[✨ You'll Know It]
 H -- No --> F
 I --> J[🪞 Reflect on Your Journey]
 J --> K[🧠 First Conversation → Now]
 K --> L[🚀 You Are In Process]
 L --> M([🌀 The Beginning])

 style Start fill:#e1f5fe,stroke:#333
 style A fill:#fff3e0,stroke:#333
 style B fill:#e8f5e9,stroke:#333
 style C fill:#fce4ec,stroke:#333
 style D fill:#f3e5f5,stroke:#333
 style E fill:#fff9c4,stroke:#333
 style F fill:#e0f7fa,stroke:#333
 style G fill:#c8e6c9,stroke:#333
 style H fill:#ffccbc,stroke:#333
 style I fill:#d1c4e9,stroke:#333
 style J fill:#b2dfdb,stroke:#333
 style K fill:#ffcdd2,stroke:#333
 style L fill:#ef9a9a,stroke:#333
 style M fill:#4caf50,stroke:#333,color:#fff

🧠 You Have the Tools

You’ve learned how to:

Appendix 06: Example CRITIC Session: A Live Demo

Fri, 28 Aug 2026 10:00:00 +0100

This is a sample transcript of a full CRITIC loop — using the AI as a coach, critic, and journal. Use it as a reference or template for your own sessions.

 sequenceDiagram
 participant U as User
 participant AI as AI (CRITIC Mode)

 rect rgb(230, 245, 255)
 Note over U,AI: 🧗 Coach Phase (Start of Session)
 AI->>U: What's the most valuable thing you could accomplish?
 U->>AI: Explain CRITIC framework clearly.
 AI->>U: If you could only do one great thing here?
 U->>AI: Make this chapter personal and transformative.
 AI->>U: Where are you holding back?
 U->>AI: Unsure how vulnerable to be.
 AI->>U: How could you challenge yourself?
 U->>AI: Stay fully present, don't jump to other sections.
 AI->>U: Summary: Aiming for clarity, emotional depth, personal truth.
 end

 rect rgb(245, 245, 220)
 Note over U,AI: 🔍 Critic Phase (Midway Reflection)
 AI->>U: What's working right now?
 U->>AI: Energy, tone, honesty.
 AI->>U: What's missing?
 U->>AI: One more concrete example.
 AI->>U: Where's the heat?
 U->>AI: The mirror—how AI reflects mental patterns.
 AI->>U: What would take this to the next level?
 U->>AI: Include a guided prompt they can try.
 AI->>U: Feedback: Structure is strong; bring the user into the loop.
 end

 rect rgb(240, 230, 255)
 Note over U,AI: 📓 Journal Phase (End of Session)
 AI->>U: Summarize what we've done.
 U->>AI: Wrote the heart of CRITIC, proved the process.
 AI->>U: What did you learn about yourself?
 U->>AI: I still hold back, but leaning in transforms the work.
 AI->>U: What questions should we revisit?
 U->>AI: How to introduce journaling without overwhelming flow.
 AI->>U: What's worth remembering?
 U->>AI: Readers need the experience, not just the explanation.
 end

 rect rgb(220, 240, 220)
 Note over U,AI: ✅ Summary
 AI->>U: CRITIC loop: Coach → Critic → Journal. Use this structure to grow.
 end

🧗 Coach Phase (Start of Session)

AI: What’s the most valuable thing you could accomplish in this session?
User: I want to clearly explain the CRITIC framework in a way that sticks with readers.

Appendix 11: Amplified Log Template

Fri, 28 Aug 2026 10:00:00 +0100

This log helps users track and reflect on their best AI-assisted sessions — the ones that actually moved something forward. The goal is to help readers identify what led to breakthroughs, so they can replicate them more often.

 flowchart TD
 Start([✨ Amplified Creation Moment]) --> A[📓 Open Amplified Log]
 A --> B[🗓️ Record Date & Session Title]
 B --> C[🎯 Note the Goal]
 C --> D[🤖 List Prompts/Techniques Used]
 D --> E[🧵 Describe What Happened]
 E --> F[💡 Capture the Breakthrough]
 F --> G[📈 Record the Result/Output]
 G --> H[🧠 Write Your Reflection]
 H --> I[🔁 Note Follow-Up Needed]
 I --> J[✅ Log Complete]
 J --> K[🔄 Review Logs Periodically<br>to Identify Recurring Patterns]

 style Start fill:#fff9c4,stroke:#333
 style A fill:#e1f5fe,stroke:#333
 style B fill:#fff3e0,stroke:#333
 style C fill:#e8f5e9,stroke:#333
 style D fill:#fce4ec,stroke:#333
 style E fill:#f3e5f5,stroke:#333
 style F fill:#fff9c4,stroke:#333
 style G fill:#e0f7fa,stroke:#333
 style H fill:#c8e6c9,stroke:#333
 style I fill:#ffccbc,stroke:#333
 style J fill:#d1c4e9,stroke:#333
 style K fill:#b2dfdb,stroke:#333

📓 Amplified Creation Log

Use this log to record your best freestyle cognition sessions — the ones that led to clear progress, unexpected insight, or creative breakthroughs.

Agents From First Principles 03: AI Agent Keeps Making the Same Mistake? Add a Critique-and-Revision Loop

Sat, 08 Aug 2026 16:23:00 +0100

An AI agent can fail in a particularly frustrating way: it produces an answer that is almost right, you ask it to improve the answer, and it produces another answer with the same underlying defect.

Sometimes the wording changes. Sometimes it adds more explanation. Sometimes it becomes longer and more confident. But the important mistake survives.

That usually means the system is doing this:

prompt
 ↓
model
 ↓
answer

or this: