Thank you. 581 of you joined this place, many back when it was OpenClaw Builders, and I don't take that lightly. Here's where we're headed. TopOfMind is about practical AI systems that actually run a business: agents built with Hermes, OpenClaw, Grok Bot and GoHighLevel that answer the phone, follow up with leads, and get local businesses found and recommended by AI. I spent years as an AWS Enterprise Support Manager watching what breaks in production, so we build with guardrails first. What's coming: - Weekly Live Build + Q&A, every Wednesday at 12 PM PT. Bring what you're building. - Cleaned-up classroom courses, starting with Quick Start. - A free safety audit I ran on a real agent network, posted here soon. - Later, a Multi-Agent Lab for builders who want to go deeper. House rules: 1. Share builds, questions and lessons, and help each other. 2. No promo spam: no unsolicited offers, affiliate links, or "DM me" pitches. 3. Write like yourself. Pasted AI text with nothing of your own gets removed. 4. Be respectful. No harassment, scams, or unwanted DMs. 5. Keep API keys, passwords and client data out of posts and screenshots. Posts that break these get removed, and repeat or spam accounts get removed from the community. Drop one line below: what's the one task you want an agent to take off your plate? — Keith

The security tool you install to protect a server can become the way in. The cPGuard breach is the textbook case. What happened On Nov 2, an attacker used compromised credentials to get into the vendor's update system and pushed a malicious script through versions 5.88.12 to 5.88.19. It grabbed password hashes, the root database config and root SSH files. Only servers running cPGuard are affected. It's sold for cPanel, Plesk and DirectAdmin, so if you or a client host a site on one of those panels, it's worth checking. Vendor notice: https://opsshield.com/notice/cpguard-update-channel-compromise.html The 10-minute check 1. Find out if cPGuard is on the box. Look in your hosting panel's installed plugins, or ask your host or web person directly. If it isn't installed, you're not affected by this one. 2. Check the version history. Did any of your servers install 5.88.12 to 5.88.19? That's the question to get answered today. 3. Update to 5.88.25 or later. That's the fix, but it's only step one. 4. Rotate everything the script could reach. Change server passwords, database logins and SSH keys. Copied credentials may still work after the cleanup, so updating alone isn't enough. 5. Write down who owns vendor updates. For every auto-updating tool on your servers, know who gets the vendor's security notices and who would act on them. Why this keeps happening I've seen this pattern from the inside: attackers go after security vendors because their access is trusted everywhere. A tool that updates itself with root access is a tool you're trusting with the keys. My read after 30 years in corporate America, the last 7 at AWS: treat every auto-updating security tool like an admin account. Drop a reply: who on your team would even know if a vendor's update went bad? — Keith

Most of us worry about what our AI tool says. In October, the real risk was the keys stored behind it. What happened Wiz counted at least 9 AI developer platforms that disclosed incidents last month, from Fireworks AI on Oct 6 to Northflank on Oct 30. The common thread: attackers going after stored secrets, especially AI API keys, which then get resold cheap. Northflank told customers to assume every stored environment variable was decrypted and to rotate all of them. Wiz also tracked two stolen enterprise AWS Bedrock keys. One made about 800K calls in 30 hours, the other about 1.7M in 13. Wiz estimates more than $100K at list price for the two. If you build anything that runs on an AI key (a website chatbot, an after-hours agent, an OpenClaw or Hermes setup), this is for you. What I saw at AWS At AWS, we scanned public code repos for exposed keys and warned customers when we found theirs. The biggest issue was a .env file, where developers keep keys, committed right into the repo. The lockdown checklist 1. Check your .gitignore. Make sure .env is in it before your next commit. If a .env file was ever committed, treat those keys as exposed and rotate them, since deleting the file later doesn't remove it from history. 2. Move keys into a managed key store. Use your cloud's secrets manager so keys never sit in a file at all. Your app reads them at runtime. 3. Set a spending cap. The bill lands on whoever owns the key. Set budget limits and alerts on every AI provider account you use. 4. Limit where a key can run. A key allowed to run anywhere is harder to throttle. Wiz found the key limited to US regions ran into regional quotas, which throttled hundreds of thousands of calls. On the key allowed global routing, only about 6% were throttled. 5. Rotate on a schedule. Pick a cadence, put it on your calendar, and rotate right away if a vendor discloses an incident. If a vendor holds your key Ask them one question: "Where is our AI key stored, does it have a spending cap, and when did you last rotate it?"

Every coding agent most of us use today depends on a big model running in someone else's cloud. This challenge asks what it would take to get that same capability offline, on consumer hardware. The task is to post-train Gemma 4, Google's open model, into a coding agent that can find its way around a real repository and draft fixes to real software engineering tasks on its own. Humans still handle architecture, logic, and review. There are two tracks: 1. Developer Agent Competition ($65K). You post-train Gemma 4 into an agent that resolves real engineering tasks. 2. Paper Track ($35K). You submit new, unpublished research, or you write up how you approached the competition. Total prizes are $100,000, and the entry deadline is November 25, 2026. This lines up with what we keep saying in here. An agent is only as useful as the loop around it: a plan before it touches files, tests it has to pass, and a human at the gate. A local agent makes that loop cheaper to run, and you don't have to send your code to a third party. A few questions for the group: - Is anyone entering? I'd like to put a small team together. - Who has done fine-tuning or reinforcement learning on an open model before? - What hardware are you running locally? Not a vendor brochure. Just a good problem. — Keith

Two AI agents launched this summer that want full access to your email, calendar, and credit card. I'd tell most business owners to wait. Instinct is the one people are raving about. You text it like a person, and it books the reservation, negotiates the bill, and follows up without being asked. It's genuinely impressive. It also had a rough first week in public: → A user disconnected Gmail and still received an inbox summary hours later → A founder emailed himself hidden instructions as a test, and the agent followed them → Another founder reported it sent an email on her behalf without checking first Meta's Muse launched last week with a more careful design. It works on its own isolated computer, can't see your passwords or payment details, and lets you require approval before it acts. It's free to start, with paid tiers at $20 and $100 a month. But it's Meta, and trust is earned over years, not in a press release. I spent years leading Enterprise Support at AWS. The question we asked before anything touched a customer's production environment was simple: What's the blast radius if this goes wrong? An agent connected to your inbox, your clients, and your card has a very large blast radius. My advice for small business owners right now: 1. Try these on a throwaway account, not your business email 2. Set approvals to "always ask" 3. Keep them away from payroll, client data, and payments for now These tools will get there. The capability is real. But capability and trustworthiness aren't the same thing, and your customers are trusting you, not your AI. Would you give an AI agent access to your business inbox today? Why or why not?