Practical notes on automation, AI systems, internal apps and the decisions behind software that people actually use.
How to automate invoice processing step by step: one inbox, AI extraction, validation rules, matching and a human exception path, plus when to buy vs build.
Read articleOur own setup: website contact forms feed one n8n workflow that notifies us in Telegram, plus a local backup copy so an outage cannot silently lose a message.
Read articleHow an n8n workflow can find suppliers, verify prices on the actual product pages, rank offers with plain code and send reminders before deadlines, so a purchasing team stops copying prices by hand.
Read articleHow a daily n8n workflow can watch public tender portals, filter tenders with a cheap AI pre-screen and a deeper analysis, and put only the relevant ones in front of a sales team.
Read articleWhat a custom warehouse and production app typically replaces, how we build one on Supabase, and the principles that keep the numbers in it trustworthy.
Read articleStarting keyword research from your own service vocabulary only surfaces demand you already expected. A problem-first approach — Problem → Solution → Use case → Outcome — finds the searches that actually lead to the same project.
Read articleSix checks that tell you whether your AI disclosure holds up, written for a company with no legal department and no AI officer.
Read articleThe high-risk deadline moved to December 2027. The transparency rules did not: they have applied since 2 August 2026. Most SMEs have it backwards.
Read articleEveryone compares GPU rental price to API price per token. That comparison is missing the column that actually decides which one is cheaper for you.
Read articleWe rented a GPU and self-hosted an open-weight model instead of calling a third-party API. Here's what actually broke, and what we'd tell anyone trying this.
Read articleHumanEval and MBPP are saturated — frontier models clear 90%+ and the score stops telling you anything. BridgeBench evaluates across seven production-relevant categories instead of one isolated task type.
Read articleFable 5.1 is Anthropic's new flagship, released September 2026 — matching or beating Opus 5 on most benchmarks while using fewer tokens. What that means for a project you're planning.
Read articleThe demo always works. The failures we get called in to fix all happen after that, in the gap between a clean example and messy real usage. Three patterns show up over and over.
Read articleNo-code tools win for the first three months. Then usage grows, and the same shortcuts that made you fast start making you slow. Here's how to tell which side of that line you're on.
Read articleMost automation projects stall the first time they meet real records. The cleanup that unblocks them takes days, not quarters, you can run it yourself, and it pays back even if you never build the automation.
Read articleReal budget ranges for a first automation project, the two costs every budget forgets, and the three numbers you can work out yourself to know whether a project is worth doing at all.
Read articleThe spike-then-collapse pattern in AI content sites is real and documented. The viral case studies illustrating it usually are not. We separated the verified policy text and published datasets from the marketing anecdotes.
Read articleA 184,000-star command-line downloader is the ingestion layer under most AI content workflows. What it actually does, why the licence is unusual, where it breaks, and the compliance questions to settle before it touches a client project.
Read articleThe category is two years old and already full of people selling the same three workflows at consulting prices. Here's what the work really involves, where it fails, and the questions that separate a partner from a demo.
Read articleA client portal removes the status-update loop between you and your customers. Here is how to work out whether it pays for itself, what belongs in version one, and the three reasons portals fail after launch.
Read articleA plain definition of an AI agent, the four levels of autonomy worth distinguishing, and an honest look at why most enterprise agent pilots die before production. Written for the person who has to approve the budget.
Read articleAn MCP server is the layer that lets an AI assistant reach your actual systems. What it contains, what it costs, where it goes wrong, and how to decide whether your company needs one — from a team running four of them.
Read articleThe whole method we use to find where a company is losing time, written out so you can run it yourself on a Thursday afternoon. No tools to buy, no consultant required.
Read articleAnthropic's new flagship doubles frontier performance while cutting costs — what it means for anyone building agentic products.
Read articleStacking another subscription rarely fixes the actual workflow problem. Here's when a small custom app beats adding one more SaaS tool to the pile.
Read articleOne repeated pattern shows up again and again in client inquiries: information trapped in one tool, needed in three others. Here's the workflow we build to fix it.
Read articleComputer-use AI agents can already operate real, undocumented internal software. That changes what "automatable" means for legacy tools and admin panels.
Read articleNo pitch deck, no generic discovery questionnaire. Here's what actually happens on the first call before we quote a fixed price.
Read articleNot velocity, not story points, not hours logged. One number tells us more about whether a project is going well than everything else combined.
Read article
Chat with us on WhatsApp