AI Weekly: The Open-Source Agent Race Heats Up — March 9-15, 2026

The open-source agent world had its biggest week yet, and Microsoft Office got a genuinely useful AI upgrade at the same time. Nous Research's Hermes Agent went from a scrappy underdog to a five-times-bigger phenomenon in a fortnight, and Anthropic quietly made Claude smart enough to remember what you're doing between two different apps. Neither story is hype for hype's sake. Both change what's realistic for a small business to automate right now.

The Deep Dive: The Agent Rivalry Escalates, and Claude Learns to Multitask

Hermes Agent goes from 22k to 110k stars in two weeks

On March 12, Nous Research shipped Hermes Agent v0.2.0, its first proper public release, and the numbers are hard to ignore: 216 merged pull requests from 63 contributors, 119 issues closed, and a jump from near-zero to 3,289 tests. The release adds a real messaging gateway (Telegram, Discord, Slack, WhatsApp, Signal, email), native MCP client support, editor integration for VS Code, Zed and JetBrains, and git worktree isolation so the agent can work on the same repo in parallel without stepping on itself.

The bigger story is the growth curve. Two weeks ago we covered Hermes Agent crossing 22,000 GitHub stars. This week it blew past 110,000 — a five-fold jump — making it the fastest-growing agent framework anyone's tracked in 2026, and a serious contender against the established players in the space.

Claude starts remembering what you're doing across apps

On March 11, Anthropic rolled shared context into Claude for Excel and Claude for PowerPoint. Build a model in a spreadsheet, switch to the slide deck, and Claude already knows what you were working on. No re-uploading the file, no re-explaining the numbers. The update also adds Skills, letting a team save a repeatable process — like turning this month's numbers into a board pack — as a one-click action anyone on a paid plan can reuse.

Why this matters:

• Hermes Agent's growth curve shows open-source agent frameworks are now genuinely competing with closed platforms on speed of shipping, not just cost.
• Rapid, unaudited growth in any open-source agent framework is worth watching closely, not adopting blind — security maturity tends to lag well behind star counts.
• Shared context between apps removes the most annoying part of using AI at work: re-explaining yourself every time you switch windows.

What This Means for Australian SMEs

If your team already uses Claude for reporting or analysis, the Excel/PowerPoint update is worth testing this week. Most SMEs lose real time re-pasting context between a spreadsheet and a deck; that's now solved if you're on a paid Claude plan.

If you've been eyeing an open-source agent for the business, hold off on picking a favourite based on hype alone. This category is moving fast enough that whatever you pick this month may look outdated by June. Evaluate on what you need it to do, not on star counts.

The practical takeaway for this week:

• Trial Claude's Skills feature on one repeatable report your team builds every month before rolling it out wider.
• Treat any agent framework with explosive growth as unproven, not just exciting. Growth and security maturity rarely arrive together.
• Keep watching this space. Two rival open-source agents shipping features this fast usually means a shakeout is coming, not a stable choice.

This kind of tool triage — working out what's actually production-ready versus what's still moving too fast to trust — is exactly what we help clients with at IntelliAgent. If you want a second opinion before you commit to an agent framework or roll Claude's Skills out across your team, get in touch.