Five AI Tools Actually Worth Watching This Week

There’s a lot of noise in AI right now. Every week brings a dozen new tools, updates, and announcements. Most of it doesn’t matter yet. Some of it does.

Here are five things from this week that are genuinely worth knowing about — explained without the jargon.

Your AI assistant can now ask a smarter friend for help

Anthropic (the company behind Claude) released something called the “advisor strategy.” In simple terms: you can now run a fast, affordable AI model for everyday tasks, and have it check in with a more capable model when it hits something tricky.

Think of it like a junior team member who handles most of the work independently but knows when to knock on the senior’s door. The junior does 95% of the job. The senior weighs in only when it matters.

What this means for businesses: If you’ve been put off by the cost of using AI for routine work, this changes the maths. You get the quality of the expensive model where it counts, without paying for it on every single interaction.

Your own notes can become your AI’s knowledge base

Andrej Karpathy — one of the most respected names in AI — has been showing how your personal notes, kept as simple text files, can work as a knowledge base for an AI assistant. No special database. No complex setup. Just well-organised notes. I have been using this approach for the last couple of months and am finding it groundbreaking.

If you use a tool like Obsidian (a note-taking app that stores everything as plain text files), you can point an AI at your notes folder and ask it questions. It reads your files, finds what’s relevant, and gives you answers grounded in your knowledge — not the internet’s.

What this means for businesses: This is the approach we recommend. Before you invest in expensive AI platforms, get your existing knowledge organised. If your team’s operational docs, processes, and decisions are written down clearly, you’re already most of the way to having an AI that understands your business.

Open-source AI agents are becoming genuinely usable

Two tools got significant updates this week:

OpenClaw is a free, open-source framework for building AI agents — software that can carry out multi-step tasks on its own. It now works with GLM 5.1, a new model designed specifically for long, complex tasks. The important part: you can run it entirely on your own computer. Your data never leaves your building.

Hermes Agent hit version 0.8, making it dramatically easier to run AI agents locally. No cloud account needed. No data sent externally. It now works smoothly with several open models and includes features like background task notifications — it tells you when it’s finished rather than making you watch.

What this means for businesses: If data privacy is a concern (and it should be), the option to run AI tools on your own hardware is becoming real — not just a technical possibility but something that actually works well enough to use. This is worth watching if you handle sensitive client information or operate in a regulated industry.

The pattern worth noticing

These tools all point in the same direction: AI is becoming cheaper, simpler, and more private. The cost is dropping. The setup is getting easier. And the option to keep everything in-house is increasingly practical.

But the technology isn’t the hard part. The hard part is knowing where to apply it — which workflows would actually benefit, and which ones just need better coordination.

That’s always where we start. Not with the tool, but with the work.