Why Agentic Coding will Always Introduce Errors
LLM-centred agentic systems will always make mistakes, just like human engineers. Human-authored tests are all that can stop an agentic quality drop.
Phroneses.com
Latest posts
Hashgraph: a Byzantine fault-tolerant state‑driven, history‑interpreting consensus algorithm
A walk‑through of how gossip communication alone produces a single global order without leaders, coordination, or messages in a Byzantine faulty distributed system.
There is no such thing as Technical Debt
Most technical debt is not debt at all. It is the unavoidable cost of changing yesterday’s decisions to support today.
Preventing Double‑Spend: Idempotency, Internal IDs, and the Discipline of State Before Side‑Effects
An idempotent service must never trust client‑supplied names. This piece shows how server‑side identity, atomic state, and UUIDv4 naming prevent double‑spend and keep distributed systems safe under failure.
Bias, Metrics and Technical Debt
Technical debt is often ignored due to bias; using clear metrics turns it into a measurable business risk, improving predictability, reducing fragility and preventing future delivery slowdown.
Retry Windows and the Limits of Idempotency
Retries hide duplicated operations your service cannot see. This piece shows why idempotency depends on the operation name you choose.
Why Early Architectural Decisions Restrict Your Future Options
Residuality is the discipline of designing boundaries that keep future options open by preventing early implementation choices from defining a whole architecture.
Software Delivery Constraints and Pipeline Throughput
Modern software delivery is not just coding. It is a seven‑stage, 32‑task pipeline. Speed gains from AI only matter when the pipeline is understood and has capacity end-to-end.
When your board wants a strategy and you have three weeks
A practical guide to producing a defensible and aligned strategy in three weeks when your board has five different meanings of the word strategy.
Measuring Reliability in the Age of AI
In a delivery system reshaped by AI, metrics are the only defence against blind risk
The Myth of Complete Specifications
A complete specification is an illusion; real systems demand incremental discovery, and SDD only works when treated as a tool inside an iterative TinySlice loop. Winston Royce knew this in 1970.
The Cost Curve of Unchecked LLM Context
LLM workflows need context, but unmanaged context growth becomes expensive and degrades results. This piece outlines how to keep usage controlled and effective.
What Python async exposes that synchronous code hides
Asynchronous IO exposes hidden waits and coupling, revealing backpressure early and making system behaviour observable.
Hiring in an AI World
AI automates code generation. Evaluate engineers on judgement, failure reasoning, and system awareness.
The Limits of Stateless LLMs
LLMs have reached a ceiling; the next gains require engineering, not scale.
Vibe Coding Is Not Engineering
LLMs generate code, but they cannot see the engineering decisions that keep systems safe.
Why Junior Engineers Matter More as AI Expands
Junior engineers evolve toward judgement, verification, and system awareness as AI absorbs the mechanical act of coding.
When Urgency is High but Progress is Slow
A clear view of why leaders feel rising ambiguity and how structured judgement restores clarity without leadership abstractions.
Before You Adopt AI in Engineering, Answer These Five Questions
Most organisations think they are maturing in AI, but their workflows tell a different story. These five questions give engineering leaders a clear, stage‑aligned way to understand their real maturity and scale AI safely.
Agents Cannot Maintain Systems: The Additive-Transformative Gap in LLM Software Delivery
LLMs can generate code, but they cannot modify or maintain systems because system‑level work requires causal reasoning, not pattern‑matching.
When Code Is Cheap, Judgement Matters More
AI lowers the cost of code, not the cost of thinking. Clarity and judgement, not speed, determine whether teams build what truly matters.
The Missing Structure Agile Cannot Fix
Agile cannot fix structural gaps; delivery depends on clear ownership, boundaries, and decision‑rights across the wider organisational network.
Designing Prompts for Modern AI Systems
Modern AI systems require structured, multi‑step prompts that guide planning, critique, and long‑context reasoning.
How AI Works
An explanation of how large language models actually function and why they should not be treated as miniature humans.
Team AI is the Next Step Beyond Cut-and-Paste AI
Individual AI delivers diminishing returns; meaningful improvement comes from strengthening the collective workflow.
The Big AI Gains Come From Teams, Not Individuals
The real gains from AI come from teams.
The Big AI Gains Come From Teams, Not Individuals
The real gains from AI come from teams.
Global AI Trends 2024–2025
Global evidence shows rapid AI adoption, rising capability, and widening gaps between regions and firms, with the US driving investment and commercial uptake.
Global AI Trends 2024–2025
Global evidence shows rapid AI adoption, rising capability, and widening gaps between regions and firms.
AI and Brands: A Practical Framework for Protecting and Strengthening Brand Equity
AI strengthens brands when it improves precision, consistency, and control — and destroys them when it introduces noise.
AI for Luxury Watchmaking: Discipline Over Display
Luxury maisons must adopt AI with restraint, using it as a precision instrument that protects craft, tone, and identity.
10 Everyday AI Workflows That Save Hours
Ten simple AI workflows that save minutes each day and compound into hours each week, helping people work more efficiently.
Building Safe, Compliant and Sustainable LLM Systems
LLM systems behave differently from traditional software and require layered safety, strong governance, observability, and architectural discipline to operate reliably and sustainably.
Evaluating AI Systems: Metrics that Matter
Evaluating AI systems means measuring real behaviour, not synthetic benchmarks.
How to Evaluate the Output of an AI Chat Session
A practical guide to assessing the quality, reliability, and safety of AI chat session outputs.
How to Use AI Safely and Effectively
Guidance on using AI safely and effectively, grounded in recent examples of misuse and emerging best practices.
Latency is architectural
Most latency comes from retrieval hops and orchestration, not the model; RAG pipelines often recreate microservice-style chatter that slows systems down.
Chat Interface to System Component
AI systems behave like probabilistic components; engineers must build structured interfaces and layered constraints to make them reliable inside software systems.
What Tech Executives Need to Know About Working With LLMs
Executives must treat LLMs as probabilistic systems requiring controls, governance, and new forms of oversight.
Transforming Your Business for AI
AI adoption is an organisational transformation requiring mandates, measurement, and redesigned processes.
What AI Is (and Isn't)
A clear explanation of what AI is—and is not—cutting through hype to
What software engineers need to know about LLMs
Engineers must think in tokens to avoid test‑to‑production mismatches.
A Beginner's Guide to AI Chatbot Prompting
Clear, practical prompting habits to help you get faster, more reliable results from everyday AI tasks.
How to Evaluate A Company's AI Claims
A framework for evaluating claims made about AI systems, focusing on evidence, capability, and verifiable performance.