Matious

Insights

Articles from the Matious engineering team on taking AI systems from pilot to production: what we build, what breaks, and how we fix it.

A scale-model railway junction where one track splits in two, with a thrown switch lever and a green signal lamp on the diverging track

Models/Nov 12, 2026/8 min read

December 1 on the Gemini API: a price rise and two forced model swaps

On December 1, Gemini 4 Argon's list price doubles and Google reroutes Gemini 2.5 Pro and 2.5 Flash to newer models. Why we handle model changes as incidents, and how to build an agent stack that switches on your schedule.

Read the article→

A sheet labeled BUILD and RUN with grey bars, one RUN bar underlined in green, beside a calculator and a pencil on a grey desk

AI costs/Nov 9, 2026/8 min read

What an AI agent really costs to build and run

Most AI agent price ranges online have no basis. What actually drives the build cost and the run cost, with a worked example at current list prices and the questions to ask a vendor.

Two smartphones, one with an AI sparkle and one with a verified-developer badge, linked to a checklist

Mobile/Oct 24, 2026/6 min read

iOS 27, Android developer verification, and what app teams should plan for

iOS 27 lets apps call any language model, on-device or in the cloud. Android now ties apps to a verified developer. What React Native teams should do now.

A small token tagged minus 90 percent beside an AI agent terminal printing a very long receipt

AI costs/Oct 21, 2026/6 min read

Cheaper tokens, pricier agents

Token prices fall while the cost per agentic workflow climbs. Where agent budgets go, and the routing, caching and measurement habits that keep them in check.

An AI shopping agent carrying a bag from a product page to the checkout, with a Black Friday tag

E-commerce/Oct 19, 2026/6 min read

Your next shopper is an agent

AI-referred shoppers now convert better than other traffic. Before Black Friday, check your store the way an agent reads it: product data, prices, checkout.

Five AI model chips in a row, from a small one tagged $0.50 to a large one tagged $20

Models/Oct 14, 2026/6 min read

A builder's guide to the autumn 2026 model wave

Three frontier launches in three weeks, priced from $0.10 to $20 per million tokens. What shipped, how to compare it fairly, and how to build for the next one.

A router sending requests evenly to identical servers, with an unplugged session store set aside

Engineering/Oct 9, 2026/6 min read

MCP goes stateless

The July 2026 MCP spec drops sessions and the initialize handshake. What breaks, what gets easier to host, and how we're approaching migration and security.

A magnifying glass over a credit-scoring dashboard with a fairness chart, beside a bank column and audit files

Regulation/Oct 6, 2026/6 min read

What French and EU financial supervisors now expect from AI

The ACPR will supervise high-risk AI from December 2027 and doubts that dropping sensitive variables makes a model fair. What French firms should build now.

A desk calendar with dates circled under a ring of EU stars, next to a checklist and an AI chip

Regulation/Sep 30, 2026/7 min read

What the EU AI Act still asks of your product after the Omnibus

The Omnibus moved the high-risk deadlines, not the transparency duties or GPAI enforcement. What applies now, what lands in 2027-2028, and how to use the gap.

A laptop with code in front of a window onto Casablanca and the Hassan II Mosque, with clocks for Casablanca, London and Paris

Nearshore/Sep 22, 2026/6 min read

Why Morocco is becoming Europe's AI engineering nearshore

A national AI plan in implementation, a data-center buildout and an offshoring sector moving into engineering. What the 2026 numbers mean for teams hiring here.

A document's name, IBAN and phone fields turning into tokens as they pass a glass shield on their way to an AI chip

Security/Sep 15, 2026/6 min read

AI is now both attacker and target

IBM finds one in four malicious breaches is now AI-enabled. How we design LLM stacks so that a breach exposes tokens instead of personal data.

A tax code book and a 1040 form linked by threads to an AI answer card with checked citations and one rejected

Tax tech/Sep 8, 2026/6 min read

AI tax research goes mainstream, and so does the hallucinated-citation problem

Most US tax professionals now use AI for research weekly, and fake citations have reached the Tax Court. What Clinco and the IRS OPR guidelines mean for tools.