Writing

Notes from the build

Articles on Laravel architecture, retrieval-augmented generation, AI agents, small language models, vector search and agentic commerce, drawn from production work rather than demos.

Page 1 of 4 · 48 articles

Hiring9 min

What to ask a full stack developer before you hire one

Marketplace hiring guides are written to get you to a hire fast: an average hourly rate, a 48 hour turnaround, a checklist of frameworks. Speed of placement is what they optimise, not whether the project survives its first year. The skills a full stack developer actually needs, the questions that show whether someone can own a system from architecture to handover, and what a good answer sounds like.

5 November 2026Read

Web Performance8 min

The SaaS landing page that survives the traffic it asks for

Search for how to build a SaaS landing page and every result is about design: hero sections, social proof, a stunning gradient. Design advice is saturated. What nobody covers is what happens on launch day, when the page works and the traffic it was built to attract arrives all at once. Core Web Vitals under load, images, analytics that do not block the page, and signup forms that do not lose leads.

4 November 2026Read

SaaS Architecture9 min

What a SaaS dashboard needs before you let customers see it

Every page ranking for building a SaaS dashboard is about how it looks: the card grid, the chart library, the Tailwind template. None of them covers what makes a dashboard trustworthy: which numbers are scoped to the tenant, how stale each one is allowed to be, what a brand new account sees, and the permission check behind every widget. The list I work through before launch, with the code and the tests.

3 November 2026Read

Agent Integration10 min

An AI agent in Microsoft Teams without rewriting the agent

Microsoft's documentation assumes you are starting from nothing: a new project in their SDK, or a new agent in Copilot Studio. Most people asking already have an agent that works on the website or in Slack and a customer who runs on Microsoft 365. How to bring that agent into Teams with a thin adapter, what the protocol actually asks of you, and the tenant-admin reality the docs gloss over.

2 November 2026Read

AI Support9 min

Build vs buy for AI support: the five questions that decide it

Every page ranking for this question is published by a company selling an AI support product, and every one concludes you should buy, from them. The honest answer is that buying is often right, and the cases where it is not are predictable. Five questions that decide it, the cost arithmetic at three volumes, and the hybrid most B2B teams end up with.

1 November 2026Read

AI Markets6 min

Has the AI bubble burst? Why AI shares are falling in China and the US

October 2026 was the worst month for AI stocks in years on both sides of the Pacific. China's AI and chip index has lost 40% since June, and the Nasdaq 100 shed more than $2 trillion in a month. The two sell-offs look alike from a distance and have quite different causes. What actually happened, why, and whether bubble is the right word.

31 October 2026Read

Security6 min

How the $6.3 billion Binance hack happened: a laptop, a signing system and one Ethereum feature

On 29 October 2026 attackers drained about $6.3 billion from Binance's hot wallets across ten blockchains, most of it in twenty minutes. No private key was stolen. What is known so far about how they got in, the EIP-7702 trick that emptied the Ethereum wallet in one transaction, where the money went, and the lesson for anyone running a system that signs things.

31 October 2026Read

AI Support13 min

Building a B2B AI support agent that knows what the customer is actually paying for

Every page ranking for AI support in B2B SaaS is a vendor comparing itself to the others, and none of them deals with the failure that matters: the right document retrieved for the wrong customer. Why a support answer depends on plan, contract and account state, how to resolve those before the model sees anything, and the tests that prove it cannot see past them.

31 October 2026Read

No-Code SaaS10 min

From Lovable to production: what a no-code SaaS rebuild actually involves

Lovable, Bolt, Replit and v0 hand you real code, which changes what a rebuild means. Most of these apps do not need rewriting from scratch: the interface usually survives and the data layer usually does not. What transfers, what gets rewritten, the five phases in order, and what each path costs in weeks and money.

29 October 2026Read

AI-Assisted Development12 min

Multi-tenancy, billing and roles: the SaaS plumbing AI coding tools still get wrong

Claude Code and Cursor will scaffold a multi-tenant Next.js SaaS with Stripe and roles in an afternoon, and it will look right. The leaks are in the plumbing: tenant IDs read from the request, subscriptions that only know two states, role checks that live in the interface. The mistakes I find most often, why agents keep making them, and the harness that catches them before a customer does.

28 October 2026Read

No-Code SaaS11 min

No-code got you to 100 users. Here is the bill for the next 1,000.

No-code was the right call for the first hundred users. The next thousand arrive with a bill in four currencies: pricing that meters your success, ceilings you find by hitting them, security nobody reviewed, and a rebuild that gets dearer every month you wait. What each one costs, what you can take with you when you leave, and the rebuild in weeks and money.

27 October 2026Read

AI-Assisted Development11 min

Build SaaS with AI: what the 3-10x speedup claims leave out

Every guide to building SaaS with AI quotes a three to ten times speedup. The controlled studies measured something much smaller, and one caught experienced developers getting slower while feeling faster. Where AI genuinely shortens a SaaS build, where review eats the gain, and the honest number for a whole platform.

25 October 2026Read