Short beats from M-Tech Labs AI — company updates on what we've launched, and industry items with our take on what they mean for SMEs and the MSPs supporting them.
Our Head of AI & Automation, Rob De Gouveia, took the M-Tech seat at an invite-only lunch in Mayfair this week hosted by Replit's CEO Amjad Masad — alongside JPMorgan, Databricks, Worldline and Travalent — to mark Replit's first international office, coming to London. Wrote up what we took away.
See the page →
Of the Insights posts we've shipped so far, this is the one readers keep messaging about. First-person, honest about the 3am pace of the shift, and the reason the rest of the series exists in the tone it does. If you only read one, start here.
See the page →
Florida's AG has opened a criminal investigation into OpenAI, claiming ChatGPT advised a Florida State University mass shooter on gun choice, ammunition, timing and where on campus the crowd would be densest. OpenAI says the model returned factual info available elsewhere online. The AG: if a human said the same thing, they'd be charged as an accessory. First criminal probe of its kind.
Read the source →
The Insights area goes live with the opening five pieces — from the emotional weight of the pace AI is moving at, through shadow-AI governance under UK GDPR, to the economics of small software now the shackles are off. Each carries a typographic hero so the stream reads as a deliberate series. More will follow; start with whichever title speaks to where you're at.
See the page →
Hobbs Recovery Services were running two collection sites off a COVID-era form. We replaced it with a purpose-built booking portal — mobile-first PWA for customers, full admin workflow for staff, single sign-on and document upload at point of booking, self-hosted on their own server. First case study we can publish, and it's up now.
See the page →
Carl Windsor, Fortinet's CISO, on what Claude Mythos means for defenders: vulnerabilities that took months to weaponise can now be weaponised in hours, and manual patch cycles don't survive that. Proactive, automated, continuous — AI as a force multiplier, not a replacement. Notable because Fortinet is the perimeter we run on ourselves.
Read the source →
M/agnet is the public Wi-Fi layer behind hospitality and public-venue networks — branded captive portals, low-friction sign-up, auditable consent, returning-guest recognition and Mailchimp sync. UniFi-first, multi-tenant, built to extend. Joins M/uster and M/esh as the third platform we can stand up for a client.
See the page →
Anthropic has launched Claude Design — prompt-to-prototype, reads your codebase and brand files, exports to Canva, PDF, PPTX and hands off to Claude Code. Figma's stock fell 7% on the news. Another category crossing the line from "specialist engagement" to "afternoon's work" — the web-design economics we flagged in "The shackles are off" just got a visible worked example.
Read the source →
Anthropic has shipped Opus 4.7 — stronger coding, higher-resolution vision, same price as 4.6, and available on Bedrock, Vertex AI and Microsoft Foundry from day one. The practical read for the sorts of businesses we work with is that the "too small to commission" category just shrank again. This is the commoditisation curve we wrote about in "The shackles are off," landing in real time.
Read the source →
We've shipped a dedicated page for clients wrestling with shadow AI — staff pasting client data into consumer ChatGPT, Claude and Gemini tenancies nobody in the organisation controls. It lays out the Microsoft Purview and Defender for Cloud Apps tandem that turns "we didn't know" into "here's our position," and what a readiness review looks like before you buy anything new.
See the page →
HR platforms don't see operations. Operations tools don't see people. We got tired of the seam where things fall through — leavers keeping fobs, laptops sitting in drawers, compliance lapsing quietly — and built M/uster: the operational platform we now run M-Tech on. Customer zero, fully dogfooded, and the case study is up.
See the page →
Anthropic has previewed Claude Mythos — a step-change in capability, especially at cybersecurity (181 working exploits on Firefox vulnerabilities versus two for Opus 4.6). Previewed, not released. It's going to Project Glasswing for defensive use on critical software, not to general availability. A frontier lab publicly withholding capability is a new shape of story, and worth planning around.
Read the source →
The three frontier labs that normally slug it out have started sharing information through the Frontier Model Forum — the industry body they co-founded in 2023 — specifically to clamp down on Chinese extraction of frontier-model behaviour. Commercial rivalry parks when the geopolitical stakes hit. Worth noticing because it's a new kind of industry-body action, not just a press release.
Read the source →Book a discovery session and we'll walk through the workflow, the systems and the shape of the solution.