enterpriseagenticsystems
Bringing deterministic agentic software engineering to enterprise teams, paired with The Software Factory running across The Lab over Tailscale & SSH on a student budget.
Beyond "Vibe Coding": Deterministic Agentic Engineering
In enterprise codebases and multi-repository microservices, ad-hoc AI chat prompting rapidly degenerates into context drift, broken contracts, and unvetted regressions. To achieve 5–10× developer velocity without sacrificing reliability, autonomous agents must be treated as a deterministic distributed system: bounded by in-repo ubiquitous language, constrained by deep module interfaces, executed in isolated git worktrees, and gated by non-negotiable verification suites.
Enterprise Architecture & Team Enablement
The four architectural pillars engineered to standardize how engineering teams and autonomous agents ship production software together.
In-Repo Context & Domain Codification
Implicit tribal team knowledge is codified into version-controlled ubiquitous language, entity glossaries, and Architectural Decision Records. New human engineers and autonomous agents share identical, durable ground truth without hallucinating domain boundaries.
Deep Modules & Blast-Radius Containment
Codebases are structured around deep modules: simple, narrow interfaces that hide extensive implementation complexity. Static boundaries and strict type contracts prevent agents from producing leaky abstractions or introducing cross-boundary coupling.
Composable Skill Workflows & Spec Execution
Requirements undergo automated Socratic stress-testing before code is written. Features are broken into orthogonal, file-isolated vertical slices executed concurrently by parallel subagents across isolated worktrees, verified against strict red-green test suites.
Developer Onboarding & Adoption (Next plc)
Developing playbooks to upskill developers at Next plc: progressing from single-turn autocomplete to structured multi-agent orchestration, establishing hard commit pre-checks (npm run check, Vitest), and maintaining strict human-in-the-loop review standards.
The Lab: Distributed Multi-Computer Compute Setup
Running the Software Factory on a student budget: linking the computers I already own (portable Linux development + desktop CUDA compute) over a private Tailscale and SSH network before spending on paid cloud credits.
The Lab — Hardware Nodes
Tailscale Private NetworkIntel i5-10400F · NVIDIA RTX 3060 Ti · 32GB DDR4 · 500GB SSD + 2TB HDD
Pop!_OS Linux · AMD Ryzen 7 7940HS · NVIDIA RTX 4050 · 16GB RAM
10+ year old x86_64 hardware scheduled for lightweight headless Linux installation
- •Tailscale Mesh & Remote SSH: Seamless cross-network SSH access established between the Pop!_OS laptop and the Windows workstation.
- •Software Factory Orchestration: Planning, building, and verifying projects across machines using Oh My Pi (OMP) and T3 Code.
- •Framework Exploration: Actively evaluating Hermes Agent and specialized autonomous subagent DAG runners.
- •Hybrid Cost-Aware Model Routing: Routing simple tasks to smaller local models (RTX 3060 Ti) while escalating complex architecture to frontier cloud models across ChatGPT, Gemini, and Copilot.
- •Persistent Background Worker Nodes: Repurposing the 2 older laptops into headless Linux agent hosts for background research and continuous repository maintenance.
- •Unified Cross-Device Orchestrator: Submitting high-level objectives from any device with automatic dispatch across machines, models, and isolated worktrees.