Jacob Langvad Nilsson's Post
Jacob Langvad Nilsson · article · 2026-04-20 source ↗ Reader 🖥️
Agent Planning ArtifactsAgentic LoopAgent Context Files (CLAUDE.md / AGENTS.md)Coding AgentsContext EngineeringData SovereigntyMulti-Agent Orchestration
Effective use of AI coding agents depends not on better prompting but on providing structured context — plans, specs, skills, hooks, and a clear understanding of the agentic loop — which is the real unlock for productive human-AI collaboration in software development.
practical · expository — A curated summary of six practitioner talks at a Claude Code meetup. The post is expository — it describes and frames each talk rather than arguing a single thesis, though it advances an implicit argument through its framing.
Structure
| Part | Role |
|---|---|
| Opening frame | States the overarching theme that unifies all six talks |
| Talk 1: The Plan is the Product (Hjorth/JFM) | Establishes that agents need plans, not prompts; introduces plan cards, behavioral specs, sub-agents |
| Talk 2: Sovereignty, Speed, and the Right Team (Soerensen/Velatir) | Adds organizational/strategic dimension — rapid scaling, data sovereignty, team composition |
| Talk 3: From Basic to Above-Average (Johansen/Morningtrain) | Provides the concrete toolkit: plans.md, specs.md, Skills, MCPs, hooks |
| Talk 4: AI Coworker, Beyond Coding (Sidou/EllypsisAI) | Extends the argument to non-developers; introduces 'folder = agent' mental model |
| Talk 5: How AI Coding Agents Actually Work (Andersen/IBM) | Demystifies the agentic loop — model, prompt, tools, while loop |
| Talk 6: Applying Generative AI Across the Software Lifecycle (Taibi/SDU) | Academic perspective — AI beyond code generation toward autonomous DevOps |
| Closing acknowledgments and community call | Establishes community context — 150 attendees, multiple sponsors, next meetup planned |
Key terms
- structure
The combination of plans, agents, skills, hooks, folders, and understanding of the agentic loop that enables effective AI coding agent use - plan cards Agent Planning Artifacts
Structured planning artifacts that define work for AI agents before prompting begins - agentic loop Agentic Loop
The while-loop architecture of AI coding agents: model + prompt + tools iterating until task completion - data sovereignty Data Sovereignty
Strategic necessity of controlling where data resides, framed as a competitive concern for fast-scaling AI-native companies - folder = agent Agent Context Files (CLAUDE.md / AGENTS.md)
The mental model that an AI agent is simply a folder with a CLAUDE.md file — no deployment infrastructure needed
Propositions
- explicitPrompting alone is not the unlock for effective AI coding agent use; structure is.conf: high
- explicitStructure in this context means plans, agents, skills, hooks, folders, and understanding the agentic loop.conf: high
- explicitAI coding agents need a plan, not just a prompt.conf: high
- explicitcontestedWork should be delegated to a coordinated team of agents rather than conducted in one big chat session.conf: highcontradicts ←Genie Lessons: Nobody Wants Agents · p1 — Beck asserts nobody wants agent swarms, people want specified outcomes — directly opposing the prescription to delegate work to a coordinated team of agents.
- implicitLeveling up an entire dev team matters more than improving a single developer's AI skills.conf: medium
- explicitThe four key mechanisms for team-wide adoption are plans.md, specs.md, Skills, MCPs, and hooks.conf: high
- implicitNon-developers who think like programmers and plan like engineers can build production AI agents.conf: medium
- explicitdevelopedAn AI agent can be created simply by defining a folder with a CLAUDE.md file — no deployment infrastructure needed.conf: highextends ←The boring skill that decides how much AI can help you · p5 — The 'house rules note' the AI reads and follows is a personal, non-coding-domain instance of the folder+CLAUDE.md-defines-an-agent pattern described in the Claude Code meetup notes.
- explicitAI coding agents are not magic — they are a while loop over model, prompt, and tools.conf: high
- implicitUnderstanding the mechanism protects against both hype and fear about AI agents.conf: high
- explicitAI in software development is moving beyond code generation toward reasoning about architectures and autonomous DevOps.conf: medium
- implicitSpeed of company scaling is a deliberate strategic choice enabled by AI tooling.conf: medium
Arguments
The key to getting value from AI coding agents is providing structured context, not crafting better prompts. strength: medium
Explicit premises:
- Six independent practitioners all converged on the same theme
- Plans, specs, skills, hooks, and folders are what make the difference
- AI coding agents are just a while loop — their behavior is shaped by the structure you give them
Implicit premises:
- Convergence among independent practitioners is evidence of a real pattern, not coincidence
- The current dominant approach to AI coding focuses too much on prompting
- Structure is learnable and transferable across teams, whereas prompting skill is individual
Inferences:
- Organizations should invest in structural scaffolding (plans, specs, CLAUDE.md) rather than prompt engineering training
- The team that understands the loop and provides structure will outperform the team with better prompts
AI agent capability extends beyond developers to anyone who can think systematically and provide structured context. strength: low
Explicit premises:
- A non-developer built a production Sales Intelligence Agent
- His method was simply a folder + CLAUDE.md
Implicit premises:
- If a non-developer can do it with structure, then the barrier is not technical skill but structured thinking
- The 'folder = agent' pattern is generalizable beyond this single case
Inferences:
- The bottleneck for AI agent adoption is structured planning ability, not programming skill
Narratives
- Velatir scaling from 4 to 20 employees in three months, targeting 98 by end of year
→ Illustrates that AI-native development enables unprecedented organizational scaling speed - Sewar Sidou (non-developer) building SIA, a Sales Intelligence Agent for Tuco Marine boat builder, using folders and CLAUDE.md
→ Demonstrates that the structure thesis extends beyond developers — anyone with systematic thinking can build production agents - Morningtrain leveling up their entire dev team using four specific mechanisms
→ Shows that the structure approach scales across teams, not just individual practitioners
Questions addressed
- ✅ What separates effective AI coding agent users from ineffective ones?
Structure (plans, specs, skills, hooks, understanding the loop) rather than prompting skill - ✅ How do you scale AI coding practices across a team, not just individuals?
Shared artifacts (plans.md, specs.md), Skills, MCPs, and hooks provide a transferable framework - ✅ Can non-developers build production AI agents?
Yes, if they think systematically and provide structured context (folder + CLAUDE.md) - ⭕ Where is AI in software development heading beyond code generation?
Pointed toward autonomous DevOps and self-adaptive systems, but details are thin
Gaps
- incompleteThe post summarizes slide decks (149 pages) that are only available as images. The textual content is a curator's frame, not the primary source material. Key details, evidence, and nuance from the actual presentations are inaccessible.
Read next: Access the original slide decks or video recordings if available to extract the full arguments from each talk. - incompleteNo failure cases or limitations of the structure-over-prompting thesis are discussed. When does more structure fail? When is ad-hoc prompting sufficient or preferable?
Read next: Look for practitioner reports on when structured approaches to AI coding agents created overhead without proportional benefit.closed byYour AGENTS.md might be a big problem — Provides the missing failure case for the structure-over-prompting thesis: instruction files degrade agent performance once they exceed the ~150-200 instruction budget, showing when more structure hurts. - incompleteThe 'data sovereignty' claim from the Velatir talk is mentioned without elaboration. Why has it become a strategic necessity specifically for AI-native fast-scaling companies?
Read next: Investigate the original Velatir presentation for the sovereignty argument, likely related to EU data regulations and LLM provider data handling. - incompleteThe post does not address how these practices interact with different AI coding tools beyond Claude Code. The entire meetup is Claude Code-specific, but the thesis about structure seems tool-agnostic.
Read next: Compare with practitioner reports from GitHub Copilot, Cursor, or other AI coding tool communities to test whether the structure thesis generalizes. - uninformedNo discussion of the cost, time investment, or organizational change management required to implement structured AI agent practices across a team.
Read next: Seek out case studies that quantify the adoption cost of plans.md/specs.md/skills workflows.closed byDORA Community Discussion - Operationalizing DORA within a large organization — That extract's gap is the absence of any discussion of the cost, time investment, and organizational change management required to roll a structured practice across a team. This talk supplies exactly that: a concrete, staged change-management playbook (Kotter's 8 steps) for institutionalizing a measurement/improvement practice across a 12,000-person organization, including the multi-year time investment and the small-team (5-person) effort model.
Related documents
- [FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero
- Agent Harness Engineering
- Agentic AI and Code Reviews
- The Anatomy of an Agent Harness
- Claude Code: Best practices for agentic coding
- Don't train the model, evolve the harness
- The Orchestrator's Tax
- How Anthropic Teams Use Claude Code
- Advanced Context Engineering for Coding Agents (ace-fca.md)
- Your AGENTS.md might be a big problem