This Week in AI

The Editor

16 Mar 2026 — 3 min read

This week’s AI news shows the technology spreading into everyday tools and specialised niches rather than just massive infrastructure bets. Smaller companies and product teams are experimenting with practical uses—from AI agents that can run locally on personal computers to startups preparing businesses for future quantum computing workflows.

At the same time, AI continues to seep deeper into specific industries such as finance, robotics, and enterprise automation. The overall trend suggests the next phase of AI development may be less about headline-grabbing scale and more about integrating AI into real workflows, devices, and specialised domains.

Perplexity experiments with AI agents that run directly on your personal computer

Perplexity is experimenting with a system called “Personal Computer,” designed to run AI agents directly on a user’s machine rather than entirely in the cloud. The project builds on the emerging OpenClaw agent framework, which allows AI models to interact with tools, files, and APIs in a structured way. By bringing OpenClaw-style agents to a local environment, Perplexity aims to let users run automated tasks on their own documents and applications.

The idea reflects a broader shift toward agentic AI operating closer to the user’s data, where models can reason about files and workflows locally instead of sending everything to remote services. While this could enable much more capable personal assistants, it also raises new questions about security boundaries, permissions, and how safely AI agents should be allowed to interact with a user’s system.

OpenAI launches GPT-5.4 for professional work

OpenAI’s biggest product move this week was the release of GPT-5.4, which it says is now available in ChatGPT, the API, and Codex, with a GPT-5.4 Pro tier for maximum performance. That makes this less of a routine model bump and more of a platform-wide upgrade aimed at knowledge work, coding, and agentic workflows.

NVIDIA turns GTC 2026 into a major push for open models and AI agents

At GTC 2026, NVIDIA moved beyond chips and leaned hard into the agent stack, announcing NemoClaw for OpenClaw and expanding its Nemotron open-model strategy. The broader signal is that NVIDIA is trying to own more of the software and runtime layer for agentic AI, not just the hardware underneath it.

Anthropic’s Pentagon fight escalates into one of AI’s biggest policy battles

Anthropic’s clash with the U.S. Defense Department kept intensifying this week after the company said it had been designated a supply-chain risk and Reuters reported it sought a stay while pursuing legal challenges. This is one of the most consequential AI policy stories right now because it sits at the intersection of military use, procurement power, model safeguards, and whether governments can effectively blacklist frontier AI vendors.

A startup is pushing enterprises to prepare for quantum computing before it arrives

While most companies still treat quantum computing as a distant research project, Finnish startup Postscriptum is trying to change that mindset. The company argues enterprises should start adapting their software architectures now so they can run quantum workloads when the technology becomes practical. Rather than building quantum hardware, the startup focuses on tools that help businesses experiment with hybrid classical-quantum workflows and simulate quantum processes in existing environments.

The approach reflects a growing idea in the AI and computing ecosystem: organisations that wait for quantum hardware to mature may be too late to benefit from it. By building quantum-ready systems today—especially in areas like optimisation, logistics, and financial modelling—companies could gain an early advantage when usable quantum processors eventually arrive.

Meta doubles down on custom AI chips

Meta officially laid out plans for four new generations of MTIA chips over the next two years to support ranking, recommendations, and generative AI workloads. That makes this a major infrastructure story: hyperscalers are no longer just buying AI capacity, they are racing to build their own silicon roadmaps to control cost, performance, and long-term platform leverage.

This Week in AI

The Editor

Read more

This Week in AI

CppDeepSeek: Building a Local-First Agent Runtime in C++

This Week in AI

From Prompt to Ad — Testing AI Video Generators