Part of the uFawkesDojo curriculum — This primer introduces the AI-Native SDLC mindset that underpins the Dojo’s “Start here” lab and the uFawkesAI template. It cites The AI-Native SDLC Playbook (Claxton, Anthropic, 2026).

Share: X · LinkedIn · Email

The problem: vibe coding doesn’t scale

Vibe coding is the default mode when developers first adopt AI coding assistants:

  • Prompt the model, accept the output, maybe tweak it, move on
  • No systematic verification beyond “it runs on my machine”
  • No shared standards for when to use AI vs. when to write code manually
  • No measurement of whether AI actually improves delivery outcomes
  • Context and decisions stay in the developer’s head (or chat history)

This works for prototypes and personal projects. It fails for team delivery:

Vibe coding symptom Delivery impact
Inconsistent AI use across the team Unreviewable PRs, hidden quality variance
No guardrails on generated code Security issues, technical debt, test gaps
“It works” replaces “it’s correct” Incidents from AI-introduced bugs
No feedback loop on AI effectiveness Can’t justify tooling cost or improve usage

The shift: agentic engineering

Agentic engineering treats AI as a capability to be engineered, not a magic wand:

  • Structured workflows: AI use follows defined patterns (intent → spec → plan → execute → verify)
  • Explicit guardrails: Policy-as-code, automated checks, human-in-the-loop gates
  • Measurable outcomes: DORA metrics track whether AI improves delivery
  • Observable agent behavior: Telemetry on what agents do, decide, and produce
  • Continuous improvement: Regular retrospectives on AI workflow effectiveness

The AI-Native SDLC Playbook (Claxton, Anthropic, 2026) frames this as six stages of maturity:

Stage Focus Key practice
1. Ad-hoc Individual experimentation Copilot, chat UI
2. Assisted Personal productivity Prompt libraries, snippets
3. Automated Team workflows Templates, CI integration
4. Orchestrated Cross-cutting orchestration Multi-agent pipelines, policy gates
5. Governed Compliance & safety Audit trails, approval gates
6. Self-improving Continuous optimization Eval-driven prompt/rule refinement

Most teams are at Stage 1–2. The uFawkes suite targets Stage 3–4 as the starting baseline for platform teams.

Why this matters for platform teams

Platform engineers don’t just use AI—they build the platform that makes AI safe and effective for the whole organization.

Platform responsibility Vibe coding approach Agentic engineering approach
CI/CD pipelines Manual AI-generated YAML Golden-path templates with policy checks
Code review Human reads AI output Automated rule evaluation + human review
Testing “AI wrote tests, good enough” Mutation testing, contract tests, eval suites
Observability None Agent telemetry, DORA metrics, eval dashboards
Security Post-hoc scanning Shift-left policy-as-code (Rego/OPA)
Onboarding “Figure it out” Runnable “Start here” lab with validation

The uFawkes path

The uFawkes suite is built for agentic engineering:

Stack Role in agentic engineering
uFawkesAI Template + devcontainer + agent harnesses (the “Start here” lab runs here)
uFawkesObs Observability plane: metrics, logs, traces, DORA dashboards for AI workflows
uFawkesPipe CI/CD + policy plane: Woodpecker, Conftest/Rego, DefectDojo, golden paths
uFawkesDevX Developer experience: Backstage catalog, Coder workspaces, Cookiecutter golden paths
fawkes Kubernetes-native graduation target (Tekton, ArgoCD, CloudNativePG)

Your first step: the “Start here” lab

The Dojo’s entry-point lab (built on uFawkesAI v2.0.0) walks you through one complete intent → spec → plan → execute → verify cycle:

  1. Generate a repo from the uFawkesAI template (pinned to v2.0.0)
  2. Open the devcontainer — all tooling pre-installed (Claude Code, OpenCode, pre-commit, evals)
  3. Write an intent for a small change
  4. Produce a spec with acceptance criteria
  5. Create a plan the planner accepts
  6. Execute and run make validate
  7. Verify the eval passes against baseline.json

This is agentic engineering in miniature: structured, verified, measurable.

Key takeaways

  1. Vibe coding is Stage 1 — necessary for learning, insufficient for delivery
  2. Agentic engineering is Stages 3–4+ — structured workflows, guardrails, measurement
  3. The platform enables the shift — golden paths, policy-as-code, observability
  4. DORA metrics are the scoreboard — if AI doesn’t move them, it’s not working
  5. Start with the lab — the “Start here” lab is your first calibrated step

Continue learning

← All guides

8 min read

Next: DORA Primer →

Get notified when new guides ship