A framework for human-AI collaboration where the human stays in control, grows through the work, and retains every lesson learned.
AI tools generate faster than humans can review. When the output exceeds what a person can meaningfully engage with, the collaboration quietly breaks: the human stops reading and starts clicking “approve.” The human in the loop becomes decorative, and the distinctive value they bring, direction, judgment, style, goes missing from the work.
Take AI Bite is a set of principles for keeping the human genuinely present in AI-assisted work. Not by slowing the AI down, but by structuring collaboration so that every delivery is sized for real engagement.
But it goes further than review sizing. Take AI Bite builds an agent ecosystem that retains your memory, experience, and decisions across sessions and projects. The ecosystem becomes your avatar: an extension of your professional self that grows with you, remembers what you learned, and carries your accumulated expertise into every new collaboration.
The Principles¶
Nine principles govern how humans and AI agents work together. Each addresses a specific failure mode in human-AI collaboration.
| Principle | Core idea |
|---|---|
| Take a Bite | Deliver only what the reviewer can chew. If they can’t redirect it, it was too much. |
| The Human Brings the Spark | AI amplifies. The human provides direction, intuition, and aesthetic judgment. |
| Earn Your Assertions | Investigate before you claim. Verify before you act. Neither side gets to assume. |
| Critical Thinking | Understand first, review second, decide third. Then challenge your own reasoning: what did I miss? What am I assuming? |
| Know Your Context | The agent manages its own resource consumption. Don’t charge ahead until overflow. |
| Match the Room | Contribute proportionally to the project’s culture and scale. |
| Own Your Process | Disclose how the work was produced. Transparency about method is a professional obligation. |
| Know What You Own | Verify licensing before deployment. Free tier does not mean free use. |
| Think Ahead | Build the map before you walk the territory. Strategy emerges from operational maturity. |
For the full framework, see DSM_6.0_AI_Collaboration_Principles_v1.0.md.
The Engine: Deliberate Systematic Methodology (DSM)¶
These principles are operationalized by DSM, a living, versioned methodology that governs the full lifecycle of human-AI collaboration: research, implementation, governance, and disclosure.
DSM is not a static document. It evolves through a hub-spoke feedback loop where every session, every project, and every practitioner’s experience feeds back into the methodology. Protocols are tested, refined, and propagated across the ecosystem. What one project discovers improves every future project.
This is what makes the avatar possible. Session transcripts capture reasoning. Checkpoints preserve milestones. Memory files retain context across sessions. Feedback flows from individual projects to the central methodology and back. The result is an ecosystem that accumulates your expertise, not just your files.
Systems Prompt Engineering¶
Most prompt engineering focuses on crafting individual prompts. Take AI Bite operates at a different level: designing, versioning, and governing entire instruction systems across an ecosystem of projects.
This is Systems Prompt Engineering, a discipline that applies project management rigor to AI instruction artifacts. DSM’s instruction ecosystem covers 7 of 10 PMP knowledge areas (scope, schedule, cost, quality, communication, risk, and integration management) through version-controlled protocols, automated feedback loops, and cross-project propagation.
The framework operates at three levels:
| Level | What it manages | Example |
|---|---|---|
| Individual | A single prompt or instruction | A system prompt, a chat message |
| System | Coordinated instructions for one project | CLAUDE.md + command files + session protocols |
| Ecosystem | Instruction architecture across projects | Hub-spoke propagation, feedback loops, mirror sync |
For the full chapter, see DSM_6.1_Systems_Prompt_Engineering_v1.0.md.
Start Here¶
Read TAKE_A_BITE.md for the short version of the founding principle. It takes two minutes and captures the core idea: someone offers you a bite of a cookie, you take a bite the size you will enjoy; too small and you won’t taste the cookie, too much and it will cause a lot of issues.
Field-Tested¶
These principles were developed and validated across real projects spanning:
Data science: exploratory analysis, feature engineering, modeling pipelines
Software engineering: ML applications, production code, test-driven development
Open source contribution: external projects in unfamiliar technology domains (Android/Kotlin), with all contributions merged upstream
Structured documentation: complex, interconnected methodology systems with thousands of cross-referenced lines
Research synthesis: multi-source analysis, competitive landscape mapping, literature review
Administrative processes: financial documentation, regulatory compliance
They are not theoretical; they emerged from daily practice with AI agents across these domains. Practitioners working on complex multi-session tasks independently recreate DSM patterns (checkpoint directories, session handoffs, decision logs) before encountering the framework. DSM formalizes behavior that emerges naturally from deliberate work.
Recent Features¶
Latest additions to the framework (click to expand)
Systems Prompt Engineering (DSM_6.1) — A full chapter naming the discipline: version-controlled instruction systems, failure mode taxonomy, practitioner maturity model, and PMP knowledge area mapping
Document modularization — All methodology documents split into slim cores with on-demand modules, reducing context consumption while preserving full coverage
Dual licensing — CC BY-SA 4.0 for methodology documentation, MIT for scripts and code
Domain-neutrality audits — All numbered DSM files reviewed and trimmed of domain-specific language, making the framework applicable to any project type
Incomplete wrap-up recovery — When a session ends unexpectedly, the next session detects the gap and reconstructs the missing summary from the archived transcript
Session configuration recommendation — Each session receives a tailored model and effort configuration based on planned work scope
Mirror repo sync — Methodology files are automatically copied to public distribution repos after changes
Branch testing requirement — Feature branches must be tested before merging, with specific test plans per backlog item
Ecosystem Path Registry — Cross-repo paths declared in a local registry, eliminating hardcoded filesystem paths
Parallel session protocol — Run isolated evaluation tasks on independent branches without interfering with the main session
See the full timeline of 84+ features → FEATURES.md
What’s Coming¶
Structural compliance retrofit — Applying document structure standards (TOC, intro paragraphs, line budgets) across all methodology files
A queryable knowledge graph that compiles human-authored methodology into a navigable, interconnected structure, making the ecosystem’s accumulated knowledge searchable across projects and sessions
Onboarding guide — A newcomer-friendly path into the framework for practitioners encountering DSM for the first time
Links¶
Website: take-ai-bite.com
Author: Alberto Diaz Durana
Vocabulary: DSM terminology reference (spoke, hub, Level 3 branch, Pre-Generation Brief gates,
@reference, and more)
License¶
This project uses dual licensing:
Methodology documentation (DSM_0 through DSM_6, guides, TAKE_A_BITE.md): CC BY-SA 4.0
Software components (scripts/, configuration files): MIT