Category
AI & Enterprise AI
LLM integration, RAG pipelines, agentic systems, intelligent automation, and AI-native architecture for regulated enterprise environments.
60 articles
Three Years of Enterprise AI — What We Got Right and Wrong
A practitioner reflection on three years of enterprise AI work — the patterns I called correctly, the calls I got wrong, and what to take from each into 2026 and beyond.
The 2026 AI Infrastructure Shift — What's Changing Underneath
The infrastructure layer for enterprise AI is shifting in 2026. New hardware, new deployment patterns, new economics. A look at what's actually different and what it means for architecture decisions.
MCP One Year In — What's Working, What Isn't
Model Context Protocol is a year into broader adoption. The standardisation has paid off in specific ways and disappointed in others. A practitioner perspective from the trenches.
Enterprise AI in 2025 — Year in Review
A second year-end reflection from the field. What stabilised, what surprised, and what's heading into 2026.
Building the 2026 AI Roadmap — A Practitioner Framework
Annual AI planning has matured into its own discipline. A framework for building the 2026 roadmap that holds up through the year, not just through the planning cycle.
Banking AI Compliance in 2025 — What Regulators Are Expecting
Banking regulators have published more specific AI expectations through 2024 and 2025. The institutions that engage with the expectations early have an easier 2026 ahead.
Open vs Closed Models — Where the Decision Sits in Late 2025
The open-vs-closed model debate has matured. Both ecosystems are credible for enterprise use. The choice in late 2025 depends on workload-specific factors, not on broad ideology.
AI Auditing and Assurance — The Discipline That's Emerging
AI auditing has moved from a theoretical concept to a real enterprise discipline through 2024 and 2025. The frameworks are codifying; the practice is becoming professional.
Voice AI in Enterprise — Crossing the Production Threshold
Voice AI has been almost-there for years. Through 2024 and into 2025, the capability and the integration patterns have moved enough that specific enterprise use cases are now production-viable.
AI in the Energy Sector — Production Patterns in 2025
Energy is a sector where AI has been pitched into operations, regulation, exploration, and trading. The patterns that have shipped to production show recognisable shape; the marketing has been mostly aspirational.
AI in Life Sciences — Production Use Cases in 2025
Life sciences combines high-volume document work, regulated decisions, and complex scientific reasoning. The AI use cases that have moved into production show recognisable shape.
AI in IT Operations — Where the Real Productivity Lands
ITSM and IT operations are document-heavy, repetitive, and high-volume — well-matched to AI augmentation. The deployments that ship share recognisable shape; the ones that stall share recognisable failure modes.
Forecasting Enterprise AI Costs — Methods That Hold Up
Annual budgeting for AI workloads is hard. The costs have multiple drivers, the usage patterns change, the technology moves. A practitioner view of forecasting methods that produce useful estimates instead of theatre.
AI in Software Engineering — Beyond the Code Completion Era
Code completion was the first wave. Agentic coding tools, AI-driven IDEs, and autonomous-bug-fix services are the second. The picture in 2025 is more nuanced than either the boosters or the sceptics suggest.
Reasoning Models in Enterprise — Where They Earn Their Cost
OpenAI o1, o3, and the reasoning-model category have changed what AI can do on multi-step problems. The enterprise use cases are real but narrower than the marketing suggests.
Migration Patterns — From Early AI Deployments to Mature Ones
Many enterprises have early AI deployments that worked enough to ship and now show their limitations. The migration from early to mature deployment is its own programme of work.
AI-Native UX Patterns — What's Settling in 2025
AI-native applications have surfaced new interaction patterns. Some are working; some are friction. A practitioner view of UX patterns settling into production AI products.
AI Governance Frameworks Codify — What's Settled in 2025
AI governance was an evolving set of internal practices a year ago. In 2025 the frameworks are codifying — internally and externally — and the patterns that work are clearer.
Agent Infrastructure Catches Up — The Production Stack in 2025
Agent infrastructure was the gap a year ago. In 2025 the stack has matured enough that production deployment is a reasonable expectation, not a research bet.
Inference Economics in 2025 — Where the Cost Curves Have Settled
The cost-per-token curves moved dramatically through 2024. Where do they sit at the start of 2025, and what does it mean for enterprise architecture decisions?
The AI-Native Architecture Pattern in 2025
AI-native applications have moved from architectural curiosity to mature pattern. A practitioner view of what the architecture looks like when it's done well, and how it differs from AI-augmented conventional applications.
Enterprise AI in 2024 — What We Learned
A year-end practitioner reflection on what changed in enterprise AI in 2024, what stayed the same, and what to take into 2025.
Reading LLM Benchmarks — A Practitioner Guide to What They Mean
Every model release comes with benchmark numbers. The numbers are easy to read and easy to misinterpret. A practitioner view of what benchmarks actually measure and how to use them for enterprise decisions.
MCP and AI Interoperability — The Standardisation That Was Missing
Model Context Protocol arrived in late 2024 as an attempted standard for AI-to-tool connections. The standardisation matters more than the protocol details for enterprise architects.
AI in Supply Chain — Where the Genuine Wins Are Landing
Supply chain AI has been a long-running marketing category. The genuinely useful applications in 2024 are narrower than the pitches but more durable.
AI in Data Engineering — Where the Workflow Actually Changes
AI assistance in data engineering is producing real productivity gains in narrow places and overhyped claims in others. A practitioner view of where data engineers should actually adopt AI in 2024.
AI in Financial Services Compliance — Where the Programmes Are Landing
Financial services compliance is a high-volume, document-heavy, audit-grade workload. AI fits well in the right places and badly in the wrong ones. A practitioner view of where the programmes are actually delivering value.
Computer Use and Browser Agents — Where the Threshold Sits
Anthropic's Computer Use, browser-control demos from OpenAI and others — the agentic-AI-controls-the-screen pattern has crossed a threshold in late 2024. What's actually production-ready is much narrower than the demos.
Long Context Windows — What Changes for Enterprise Workloads
Million-token context windows are now commercially available. They change the design of some workloads materially, change others not at all, and introduce new failure modes worth understanding.
Self-Hosting Open LLMs in Enterprise — When It's Worth It
Self-hosting open models has gone from a research exercise to a real enterprise option in 2024. The cases where it earns its operational cost are clearer than they were a year ago.
AI Vendor Selection and Procurement for Enterprise
AI vendors are pitching every enterprise. The procurement process for AI tools needs to evaluate things conventional software procurement doesn't — model lineage, data handling, evaluation methodology, exit strategy.
Building an AI Centre of Excellence — What Actually Works
Every enterprise has an AI Centre of Excellence on the org chart or planned for one. The shape that compounds value differs from the consultancy-recommended default.
Real-Time AI vs Batch AI — Choosing the Right Latency Profile
The default is real-time. The right choice is often batch. A practitioner view of when each pattern earns its complexity, and how to design for the latency profile your workload actually needs.
Text-to-SQL Beyond Demos — What Production Deployments Actually Require
Natural-language-to-SQL has been a research demo for two decades. Current models make it credible. Making it production-grade in an enterprise data warehouse requires more than the demo suggests.
AI Systems and Enterprise Identity — Where Most Deployments Cut Corners
Authentication and authorisation are conventional enterprise architecture topics. In AI systems they tend to be deferred, abbreviated, or wired up wrongly. A practitioner view of the patterns that actually hold up.
LLM Security — Threats, Mitigations, and What Enterprise Teams Should Actually Do
The LLM security landscape in mid-2024 has more named threats than mature mitigations. A practitioner view of which threats deserve attention and which technical and operational controls actually reduce risk.
From AI Pilot to Production — The Playbook That Bridges the Gap
Every enterprise has AI pilots. Far fewer have AI in production. The bridge between the two is more about organisational discipline than technical capability. A practitioner playbook.
The Practical State of AI Agents in Mid-2024
The agent conversation has moved from hype to deployment in some categories and remains hype in others. A practitioner snapshot of where agents are actually working and where they are still demos.
Multimodal AI in the Enterprise — Where Vision Plus Text Earns Its Cost
GPT-4o, Claude 3, Gemini 1.5 brought capable multimodal models to the enterprise. The use cases that justify the cost are narrower than the demos suggest, but the ones that do justify it are worth investing in.
AI Code Assistants in Enterprise — What's Actually Shipping
GitHub Copilot rolled out broadly; Cursor and similar editors emerged; competitive options from Anthropic and Codeium gained ground. The enterprise picture for AI-assisted development in mid-2024 is more nuanced than the productivity claims suggest.
LLMOps Maturity — A Practitioner's Maturity Model
Most enterprises are operating LLM workloads on engineering intuition alone. A maturity model helps locate where you are, what to invest in next, and what the next stage actually requires.
AI in Customer Support — Where the Wins Actually Land
Customer support is the most-attempted enterprise AI use case. Most attempts produce modest results. A practitioner view of where the wins actually land — and where the productivity claims fall apart in production.
Knowledge Graphs and RAG — Two Patterns That Belong Together
Pure vector retrieval has a ceiling on enterprise knowledge. Combining it with a structured knowledge graph layer breaks past that ceiling for many real workloads.
Red Teaming Enterprise AI Systems — A Practitioner Playbook
Most enterprise AI systems are deployed without serious adversarial testing. The teams that ship with confidence are the ones that have tried to break their own system before users or attackers do.
The Case for Smaller Models in Enterprise AI
The default of routing everything to the largest frontier model is a habit, not a strategy. Open and smaller commercial models have closed enough of the gap that the case for using them is now strong for many enterprise workloads.
LLM Cost Discipline — Engineering Practices That Keep Bills Predictable
Most teams discover LLM cost through the bill. By then, the cost shape is set and hard to change. The engineering practices that keep costs predictable are not exotic, but they have to be in place from the start.
Fine-Tuning vs Prompting — How to Decide for Enterprise Workloads
The fine-tuning question keeps coming up in enterprise AI conversations. A practitioner framework for deciding when fine-tuning is worth it, when prompting is sufficient, and when retrieval is the actual answer.
Function Calling — Production Patterns for Enterprise
Function calling turned LLMs from text producers into action takers. The production patterns are constrained: a tight function catalogue, careful permission modelling, robust argument validation, and explicit human checkpoints for irreversible actions.
LLM Evaluation — The Engineering Discipline Most Teams Skip
Without evaluation, every change to an LLM system is a guess. Teams that build evaluation discipline ship with confidence; teams that skip it operate on intuition until production incidents force the issue.
Multi-Agent Orchestration — Hype Versus Production Reality
Multi-agent frameworks dominate the AI engineering conversation right now. The patterns that actually ship are narrower, more bounded, and more boring than the demos suggest.
Conversational BI — Patterns That Survive in Production
Conversational interfaces over enterprise data are tempting and easy to demo. The patterns that survive enterprise governance, accuracy expectations, and data complexity are a much narrower set than the demos suggest.
AI in Regulatory Workflows — A Production Walkthrough
Regulatory workflows are where the demand for AI augmentation is highest and where the bar for production deployment is steepest. A practitioner walkthrough of what actually ships in this category.
Prompt Engineering for Enterprise Integration Workloads
Prompt engineering for chat is one discipline. Prompt engineering for enterprise integration is another. The patterns that produce reliable structured output at scale are not the patterns that produce engaging chat.
AI-Native vs AI-Bolted-On — A Design Distinction That Matters
Adding an AI feature is not the same thing as building an AI-native application. The distinction shows up in the architecture and in the user experience — sometimes a year after launch.
AI Governance and Guardrails for Production Systems
Most enterprises talk about AI governance after the first incident. The teams that do it from day one ship faster, not slower — the discipline matters as much as the model.
Intelligent Document Processing — From OCR to Understanding
Intelligent document processing has changed shape in the last eighteen months. A practitioner view of where the real work sits when LLMs join the pipeline — and why parsing still matters more than the model.
Vector Databases for Enterprise Search
Vector databases are the easy part to demo and the hard part to run at enterprise scale. A practitioner view of the choices that actually matter when picking and operating one in a regulated estate.
The Enterprise AI Stack — A Reference Architecture
Most enterprise AI teams are assembling the same stack from the same parts. A clean reference architecture for the layers that compose an AI-augmented enterprise platform — and the design decisions at each layer.
RAG Architecture — From Demo to Production
Retrieval-augmented generation is the dominant enterprise LLM pattern of the year. The demos are cheap; the production systems are not. A practitioner walkthrough of where the work actually sits.
LLM Integration Patterns for Enterprise Applications
Most LLM proofs of concept work in a notebook and break in production. The patterns that survive deployment are not exotic — they're the ones built on enterprise integration discipline most teams already have.