Blog

Guides, products &
practical AI.

Deep-dives on the tools we build, and hands-on takes on AI and cloud — written to be genuinely useful.

FeaturedProduct

Conduit: move files straight between clouds, no downloads, no waiting

Moving data between Google Drive, S3 and other clouds usually means downloading gigabytes and uploading them all over again. Conduit skips that. It moves files straight from one cloud to the other, so terabytes transfer in the background while you get on with your day.

Read article 6 min read
AI

Advanced RAG: what to add when basic retrieval isn't good enough

Naive RAG is easy to build and easy to outgrow. It embeds your question, grabs the nearest chunks, and hopes. When that starts returning noise, these are the upgrades that fix it: query rewriting, hybrid search, reranking, and when you need them, GraphRAG and agentic retrieval. Here's what each one does and when it's worth adding.

Read 9 min read
Sustainability

The environmental cost of AI: energy, water, and the green-tech response

Every AI answer runs on a physical machine in a building that draws power and drinks water. As AI scales, so does that footprint, and the numbers are getting hard to ignore. Here's an honest look at what AI actually costs the planet, what green tech can and can't fix, and what it means for how we build.

Read 8 min read
Future of Work

The future of human-AI collaboration in the workplace

The scary headline is that AI takes your job. The boring reality unfolding in 2026 is stranger and more useful: AI takes the repetitive middle of your job, and the work that's left is more human, not less. Here's what the shift actually looks like, backed by the numbers.

Read 8 min read
Security

Generative AI in cybersecurity: defending against AI-driven attacks

The email that fooled your finance team had no typos. The voice on the call sounded exactly like your CEO. Generative AI has made attacks cheaper, faster, and frighteningly convincing. Here's how the new attacks work, and the defenses, human and technical, that actually hold up.

Read 8 min read
AI

How local LLMs work: running open-source models on your laptop

You don't need a data centre or an API key to run a capable language model. A decent laptop and one free tool will do it, fully offline, with none of your data leaving the machine. Here's how local LLMs actually work, what hardware you need, and how to get one running this afternoon.

Read 9 min read
Legal

How legal teams are using AI for contract analysis and compliance

Reviewing contracts is slow, expensive, and exactly the kind of pattern-heavy work AI is good at. Legal teams in 2026 are using it to read every page in seconds, flag risky clauses, and check compliance in real time, while keeping the judgment where it belongs. Here's how it works, how accurate it really is, and where the human stays in charge.

Read 8 min read
AI

Open-source vs closed AI models: which should you actually build on?

For a couple of years the answer was easy: the closed models were simply better. In 2026 that lead has nearly vanished, and the choice became an interesting trade-off instead of a foregone conclusion. Here's the real state of the gap, what each side is genuinely better at, and why most serious teams end up using both.

Read 9 min read
Security

Prompt injection: what it is and how to actually prevent it

It's the number-one security risk for AI applications, and there's no clean fix. A hidden line of text in a web page or email can hijack your AI agent and make it act against you. Here's how prompt injection works, why even the best models are still vulnerable, and the layered defenses that keep the damage small.

Read 9 min read
Business

How small businesses can implement AI automation to cut costs

You don't need a data team or a big budget to get real savings from AI. The wins for small businesses are unglamorous and specific: email, scheduling, invoices, first-response support. Here's what to automate first, what it costs, and how fast it pays back.

Read 8 min read
Research

SQL-of-Thought: teaching AI to write correct SQL from plain English

Turning a plain-English question into working SQL sounds easy until the query is wrong in a way nobody notices. A 2025 paper called SQL-of-Thought fixes this with a clever idea: don't write the query in one shot, plan it like a person would, and give the model a structured way to fix its own mistakes. Here's how it works and how to apply it.

Read 9 min read
AI

How to write a CLAUDE.md (and skills) that actually make AI useful

A coding agent that starts every session blank will keep making the same mistakes. CLAUDE.md and Agent Skills are how you give it lasting memory, but only if you write them well. Here's what goes in each, the progressive-disclosure trick behind skills, and a template you can copy.

Read 9 min read
AI

Prompt engineering vs context engineering: what changed and why it matters

For two years the hot skill was writing the perfect prompt. Then the job quietly changed. The wording of your message turns out to be one small slice of what the model sees, and the real craft is now curating the whole window. Here's the shift, in plain terms, with the tactics that actually move the needle.

Read 8 min read
AI

RAG vs fine-tuning vs long context: which one do you actually need?

Three ways to get a model to work with your data, endlessly confused for one another. The trick is a single question: is this a knowledge problem, a behaviour problem, or does it just need to fit in the window? Here's a clear decision guide, real cost ranges, and why most teams end up using more than one.

Read 9 min read
Data & AI

Semantics vs ontology vs context engine: what they are and when to use each

Three words that get thrown around like they mean the same thing, and don't. Here's what a semantic layer, an ontology, and a context engine each actually do, when to reach for which, and the out-of-the-box tools from Databricks, AWS and Google that ship each one.

Read 9 min read
Engineering

Spec-driven development: writing the spec before the code with AI

Vibe coding is fun until the AI confidently builds the wrong thing. Spec-driven development flips the order: you and the agent agree on a written spec first, then let it generate code against that. Here's how it works, why GitHub Spec Kit made it take off, and when to actually use it.

Read 8 min read
Opinion

AI detectors don't work, and the em-dash panic proves it

One popular detector once decided the US Constitution was written by AI. That should tell you how much to trust the confident little percentage these tools hand you. Here's why AI writing detectors fail, who they hurt, and the question we should be asking instead.

Read 5 min read
Writing

How to make AI-written text actually sound human

Everyone can smell AI writing now. The fix isn't a magic 'humanizer' tool, it's knowing the tells, feeding the model your own voice, and doing the last ten percent by hand. Here's what actually works.

Read 6 min read

Want to write for us?

Anyone can submit a post. Share a draft on AI, building software, or the future of work. Every submission is reviewed by our editor, and the good ones get published with your name on them.

Submit a post
Limited engagements each quarter

Give your business the AI edge — trained, or built for you.

Book a 30-minute discovery call. We'll assess your needs, recommend the right program or solution, and send a proposal within 5 business days.

Book a discovery callExplore AI products