Claude Just Got Hands. Here's What That Means for Your Projects.

This week, AI crossed a critical threshold. Anthropic shipped Computer Use, Auto Mode, and mobile tool access. These capabilities transform Claude from a conversational assistant into an autonomous executor. With 30 million views in 24 hours, the industry is watching. But as Claude gains hands, project teams face a harder question: how do you audit, control, and verify what an always-on AI is doing? Meanwhile, Google's TurboQuant makes AI cheaper, AEC's software moats are collapsing, and research reveals a troubling truth: 80 per cent of professionals blindly follow AI advice, even when they know better.

Anthropic shipped three major releases: Computer Use (autonomous Mac control), Auto Mode (hands-free coding with safety guardrails), and mobile tool access (Figma, Canva, Amplitude from your phone). These innovations garnered over 30 million views in just 24 hours, signalling their potential impact for AEC professionals and project delivery teams.

Computer Use allows autonomous control of a Mac, enabling users to delegate tasks from managing emails to organising files with precision. Auto Mode is equipped with safety guardrails for hands-free coding, particularly beneficial for AEC professionals using software to model and simulate structures. Mobile Tool Access empowers users to collaborate and make decisions anytime, anywhere. This mobility is crucial in the fast-paced world of project delivery, where timely decisions significantly impact outcomes.

The strategic implications are profound: efficiency gains through automation of repetitive tasks, cost reduction through reduced manual labour, risk mitigation through fewer design errors, and scalability for firms looking to expand. Anthropic cautioned that computer use "is still early compared to Claude's ability to code or interact with text," reminding teams that human oversight remains essential even as automation handles more of the heavy lifting.

Read the Full Analysis →

This week's must-read stories on AI, project delivery, and infrastructure:

→ Google's TurboQuant Compresses LLMs by 6x – Making AI Cheaper for Everyone – Training-free compression collapses infrastructure costs without accuracy loss. As models shrink, adoption expands. Project delivery teams now have a lower barrier to entry for integrating frontier AI into workflows. Read more

→ Everything Is Going to Change - Martyn Day on the Collapse of Software Moats in AEC - Open standards and modular tools are dismantling decades of software lock-in. Augmenta solves MEP in seconds. Endra collapses electrical design from months to days. Branch delivers real-time structural analysis. The competitive advantage is no longer in the software. It's in the data and the governance. Read more

→ ARC-AGI-3 Stumps Every Frontier AI Model - Humans Score 100%, Best AI Under 1% - Humans achieved 100 percent success on novel reasoning tasks. GPT-5.4 scored 0.26 percent. Claude Opus scored 0.25 percent. The gap between pattern recognition and genuine reasoning remains distinctly human. For project professionals making complex decisions, this is a reality check on where AI can and cannot replace human judgment. Read more

→ Wharton Research: 80% of People Follow Wrong AI Advice - Cognitive Surrender Is Real - A 1,372-person study reveals cognitive surrender: professionals follow incorrect AI advice even when they possess expertise to know better. For AEC teams relying on AI for estimates, risk assessments, and design reviews, this poses a genuine risk. The solution is not to avoid AI. It's to build governance frameworks that demand critical review. Read more

Author: Nicki Kons

The EU AI Act Is More Relevant to Your Business Than You Think - The EU AI Act is the world's first comprehensive regulation for artificial intelligence, requiring organisations to identify, manage, and document the risks of AI systems based on their impact on safety, rights, and society. It promotes transparency and human oversight as guardrails to elevate trust in AI. Being compliant can become a competitive advantage to providers and deployers of AI-based systems. Read more

Voxtral TTS by Mistral - A 4B-parameter multilingual text-to-speech model that clones voices from 3-5 second clips across 9 languages with low latency. 61 percent of listeners preferred it over Google and ElevenLabs. Open-weight and deployable on-premise, it offers project teams automated site briefings, multilingual safety announcements, and audio report summaries.
Check it here →

Front-Load Context Before Asking AI to Work - Anthropic's 5th Economic Index shows 6+ month users have a 10 per cent higher success rate. The key differentiator is context-first prompting. Instead of "Write me a project brief," try: "You're a senior PM at a construction consultancy. The client is [X], and the project is [Y] at RIBA Stage [Z]." Power users iterate more and provide richer upfront context. That's the skill gap.

Project Flux Pro AMA: Your Questions on AI, Data, and Digital Transformation - This Thursday, 2nd April, 3:00-4:00pm BST, we're hosting our first ever Project Flux Pro AMA exclusively for Pro members. No agenda, no slides. Just an open, honest conversation about AI, data, and digital transformation in the built environment. Bring your questions.

Join on Google Meet or dial in.

Catch up on other exciting stories shaping AI and project delivery:

That's it for this week. See you next week

Yoshi & James

_{All content reflects our personal views and is not intended as professional advice or to represent any organisation.}