Claude and Anthropic · mcp
Hosted vs Self-Hosted MCP Server for Industrial Systems
Hosted MCP server or self-hosted? Makini, CData Connect AI, vendor servers and custom builds compared on identity, KRITIS, write governance and upkeep.
9 min read
Author · CEO, Pexon Group
Phillip Pham, CEO of Pexon Group
Phillip Pham signs posts on the Pexon engineering blog as CEO of Pexon Group, across all four business areas: Claude in the enterprise, company brain, data foundation and private AI. This page lists every published post under that byline, newest first, with its area, date and measured reading time.
45 published posts
Everything Phillip Pham has published
Newest first. Business areas: Claude and Anthropic, Data foundation, Company brain, Private AI.
Claude and Anthropic · mcp
MES Integration with AI Agents: Batch Release Across MES, LIMS, QMS
MES integration with AI agents for batch release and deviation triage across MES, LIMS and QMS: the architecture, where it breaks, what GxP leaves to people.
10 min read
Claude and Anthropic · mcp
Windchill REST API for AI Agents: 4 Places It Breaks
The Windchill REST API can serve AI agents, but four things break a naive MCP server: one shared account, default config specs, POST reads and mixed write tools.
11 min read
Claude and Anthropic · mcp
Teamcenter API for AI Agents: Four Traps in SOA and JSON REST
The Teamcenter API for AI agents: which surface to build on, the five calls behind one BOM read, and four traps to close before an MCP server goes near production.
9 min read
Data foundation · source systems
CRM quote questions: read the quote, do not accept it
CRM quote questions an agent may answer: which quotation is open, its status, the customer. Accepting it, or turning it into an order, stays a click in the CRM.
5 min read
Data foundation · source systems
CRM sales questions an agent may ask the pipeline
CRM sales questions with a documented read: open deals, the owner, the next date. Stage changes, sends and deletes are not questions. They are edits.
5 min read
Data foundation · source systems
DATEV API: what an agent may read, and may not post
The DATEV API splits in two: DATEVconnect on the workstation for account postings, and the cloud gateway for accounting sequences. An agent may read, not post.
6 min read
Data foundation · source systems
Utility billing integration: what an agent may read
Utility billing integration for a Stadtwerk: the agent reads a contract on Schleupen.CS under the caller's login. Market messages stay in the billing system.
9 min read
Data foundation · source systems
Energy data management: what an agent may not send
Energy data management at a Stadtwerk sits in BelVis, Robotron or SOPTIM. An agent may read a schedule status. Sending that schedule stays inside the system.
8 min read
Data foundation · source systems
Utility network documentation: GIS, Lovion and SAP PM
Utility network documentation at a Stadtwerk lives in Smallworld, Lovion and SAP PM. An agent may ask which meters an outage covers. It does not edit the model.
8 min read
Claude and Anthropic · mcp
ERP PLM Integration with AI Agents: Four Places the BOM Breaks
ERP PLM integration with an AI agent: two MCP servers, one eBOM-to-mBOM comparison tool, and four configuration mismatches that make an agent compare the wrong BOMs.
10 min read
Claude and Anthropic · mcp
Dynamics 365 MCP Cost: What Claude Pays per Tool Call
Dynamics 365 MCP cost for Claude: 0.1 Copilot Credits per tool call to Microsoft, then the token bill, which is usually the larger line. Worked out in euros.
8 min read
Company brain · agentic ai
Custom engine agents: when Copilot's model is not enough
Custom engine agents bring your own orchestrator and models into Microsoft 365 Copilot. Stay declarative until the workflow leaves Copilot's loop.
6 min read
Company brain · agentic ai
Copilot Studio implementation: credits, layers, stop rules
Copilot Studio implementation needs Copilot Credits, a 125% stop rule, and four layers before any production channel opens — not a demo agent.
7 min read
Claude and Anthropic · mcp
Industrial MCP Server Security: The Vendor Default Is Not Safe
Industrial MCP server security, checked against five vendor OT servers: which ship write tools, what the defaults are, and what a read-only deployment needs.
8 min read
Claude and Anthropic · claude code
Claude Code audit trail: Claude API audit vs local disk
Claude Code audit trail: a Claude API audit uses the Compliance API, not the Activity Feed. Transcripts stay six years. Bedrock, ZDR and ~/.claude stay yours.
9 min read · updated 6 Oct 2026
Claude and Anthropic · mcp
MCP Server for Enterprise Systems: Build, Secure, Operate
MCP servers connect AI assistants to CRM, ERP and databases through one protocol. The enterprise angle: architecture, registry, tool poisoning, and secure setup.
8 min read
Data foundation · delivery model
Buying Forward Deployed Engineering: Palantir's Model, Buyer Side
Buying forward deployed engineering means testing whether the work lives on the vendor's platform or in your SAP, MES and PLM, not copying Palantir's GTM.
9 min read
Company brain · rag
RAG vs Fine-Tuning vs Prompt Engineering: The Decision Matrix
RAG, fine-tuning and prompt engineering solve different problems. The three-way matrix shows which fits current data, stable patterns or plain instructions.
7 min read
Claude and Anthropic · cost
Claude Code Cost Per Developer: Instrument These 8 Metrics
Claude Code exports 8 OpenTelemetry metrics. What to instrument, how to attribute spend to a team, and which of those numbers will mislead your board.
15 min read
Company brain · rag
How to Build a RAG System: The Complete Guide
RAG system build guide: chunking strategies, embedding models, vector search, retrieval quality, evaluation, and the rollout that survives a security review.
9 min read
Company brain · document processing
Intelligent Document Processing: Extract, Validate, Then Post
Intelligent document processing: extract invoices, delivery notes and POs, validate against ERP master data, then human review on exceptions.
7 min read
Private AI · inference serving
llm-d: Distributed Inference on Kubernetes
llm-d is the Kubernetes-native stack above vLLM: cache-aware routing, optional prefill/decode split, founded by Red Hat, Google, IBM, CoreWeave and NVIDIA.
5 min read
Private AI · open weight serving
Nemotron: NVIDIA Open-Weight Models for Private AI
Nemotron is NVIDIA's open-weight family. Nemotron 3 Ultra is 550B/55B MoE, OpenMDW 1.1, self-hostable when prompts must stay on your GPUs.
5 min read
Claude and Anthropic · coding agents
Claude Code vs Codex: Speed, Cost and the Orchestration Layer
Claude Code vs Codex on the same task: 11x faster, 6.6x cheaper for one, deeper testing and architecture for the other. Orchestration decides.
7 min read
Claude and Anthropic · coding agents
OpenCode: The Open-Source Coding Agent Challenging Claude Code
OpenCode is the open-source terminal coding agent with bring-your-own-model: custom agents, local models, and a real challenge to Claude Code.
9 min read
Claude and Anthropic · cost
GPT-5.6 Sol vs Claude Fable 5: The Manager and the Worker
GPT-5.6 Sol vs Claude Fable 5 in real work, not benchmarks: Fable is the manager, Sol the worker — and model routing is what decides your AI costs.
7 min read
Company brain · agentic ai
Agent Engineering: 7 Production Skills Beyond Prompting
Prompt engineering is the recipe; agent engineering is the cook. The 7 skills — system design, tool contracts, retrieval, reliability, security, metrics.
6 min read
Company brain · agentic ai
AI Workflows vs AI Agents: Who Decides — and When It Matters
AI workflows follow a fixed path a human defines; AI agents let the model decide. RAG is a workflow, ReAct is the agent pattern. When each is the right call.
6 min read
Company brain · agentic ai
Scaling AI Agents: Multi-Agent Design Is the Answer
Scaling agents is not about more infrastructure. Errors propagate, costs grow nonlinearly, and one agent owning everything becomes the bottleneck.
5 min read
Private AI · model routing
GLM 5.3 Flash: Frontier Behaviour at a Flash Price
GLM 5.3 Flash: open-weight, 1M context, and an audit of 1,000+ PRs for 12 cents where a frontier model costs over $100. Routing just got cheaper.
6 min read
Private AI · agent orchestration
LangGraph: Building Reliable Agents as Graphs, Not Free Loops
LangGraph structures agentic AI as a directed graph of nodes, edges and state — control, visibility and human-in-the-loop for production agents.
6 min read
Private AI · inference serving
LLM Self-Hosting on Kubernetes: Running Your Own Inference Platform
LLM self hosting on Kubernetes is production reality: 650B tokens a month, 30 tenants, shared GPUs. The four optimizations that make it economical.
6 min read
Private AI · agent orchestration
AI Model vs Agentic Harness: What Actually Drives Performance
The same model can shine in one harness and stall in another. Most recent gains came from the harness — tools, memory and the agentic loop.
6 min read
Private AI · open source platform
Open-Source AI Platform Architecture: The Harness Is the Platform
The open-source AI platform is the harness, not the model. How open harnesses, model routing and a 90/10 cost split keep your architecture independent.
7 min read
Private AI · llm gateway
What is LiteLLM? The Open-Source LLM Gateway Explained
What is LiteLLM? The open-source gateway that puts one API in front of every LLM provider — virtual keys, team budgets, usage tracking, no lock-in.
8 min read
Claude and Anthropic · eu hosting and gdpr
Claude GDPR Checklist: 14 Checks Before You Roll Out
There is no GDPR switch in Claude. Fourteen checks decide whether a rollout is lawful — contract, processor, retention, transfers, records — and who owns each.
15 min read
Claude and Anthropic · coding agents
Vibe Coding Explained: Where It Works and Where It Stops
Vibe coding means describing software to an AI instead of writing it line by line. The opportunities, the limits, and the review that keeps it out of production.
7 min read
Claude and Anthropic · cost
Claude API Cost — the Token Math Behind a Real Bill
Claude API cost: input, output and cache pricing per model, the token multiplier that catches teams, and how to forecast a bill from your actual traffic.
5 min read
Private AI · llm gateway
LiteLLM Proxy: virtual keys, budgets, EU control
LiteLLM proxy: Docker first, virtual keys not the master key, budgets per team. Self-hosted for control, not cheaper tokens. Monitor before the first incident.
9 min read
Claude and Anthropic · claude code
Claude Code vs GitHub Copilot: cost, tenant, audit
Claude Code vs GitHub Copilot: Copilot wins central policy, Claude wins per-developer cost. Own tenant vs GitHub's, 10% residency premium, 180-day vs six-year audit.
21 min read
Claude and Anthropic · claude code
Claude Code pricing: the 2,500 euro bill explained
Parallel terminals and night agents, not the licence, drive Claude Code pricing to 2,500 euros a month — about 1.5 senior engineers. How IT leaders budget it.
7 min read
Claude and Anthropic · claude code
When Not to Use AI Agents, and What Wins Instead
When not to use AI agents: four problem shapes where a script, a form or a person wins, and the one question that separates them from the cases agents suit.
5 min read
Claude and Anthropic · cost
LLM Vendor Lock-In: The 30% Your Price Comparison Misses
Switching model vendors is not an API swap. Anthropic's own tokenizer change makes identical text cost 30% more tokens, and that is the cheap part.
13 min read
Claude and Anthropic
Claude EU Data Residency: Which Endpoints Keep Data in the EU
Claude EU data residency, endpoint by endpoint: only Bedrock and Vertex AI keep inference in Europe today. What Foundry and the Anthropic API do instead.
13 min read · updated 21 Sept 2026
Next step
Talk to the people who write this
Thirty minutes with an architect from the team, about your systems rather than ours.