About
I'm the founder of Convalytics LLC, a software company dedicated to building innovative…
Activity
2K followers
-
Jason Green shared thisAnthropic's Claude Fable 5 was shut down for everyone last night because the government deemed it too dangerous. Does your company have a disaster recovery plan that includes the government blocking access to AI? If access to AI was blocked entirely, would your employees quickly revert back to doing their jobs by hand? Could they? I recommend at least discussing a backup plan.
-
Jason Green shared thisI think we're getting closer to how AI will be used in real-world business. Specialized AI workers. A few months ago, we learned about Ralph Wiggum loops which allowed AI to work a problem/task through to completion without constant approvals from the user. OpenClaw took that a bit further and allowed us to define separate agents with their own workspaces and skills. That was when I first had this epiphany. Create an "agent" specific to a domain and only give it the skills and context that is relevant. We could essentially define an "employee" and give them all of the "skills" they need to do the job. Whether to manage a sales pipeline, do engineering work, or to check your network for security vulnerabilities. In my case, I provisioned my agents with their own accounts for all of the websites or services they might need to access. They have their own email and github accounts. That way, I reduce the blast radius. They can't delete all of my personal emails or wreck my github repos. I treat them like a junior employee and give them the same types of guardrails. They can pull from my repos, but they don't have permission to commit to the main branch. Just like any employee, they need to create their own branch and open a pull request which I then approve when everything looks good. This is basically what OpenAI is doing with their "plugins". A plugin bundles together several "apps" which are connections to other tools and services (gmail, google drive, github...), along with "skills" which are text files that provide context for a more specific task (build a dashboard, validate data, write requirements docs...). I'm not a fan of the name "plugins" because these are much more like standalone agents as in our OpenClaw example. OpenAI likes to call them "packaged capabilities for Codex". Naming things is hard. The point is, this is how I'm viewing everything I do these days. If I'm going to do a task more than once, maybe it makes sense to create a plugin/agent with just the right context to let AI handle it for me the next time. And yes, we're doing the meme. There is a lot of setup involved, but I believe that once we organize around some standards, this will go much more efficiently.
-
Jason Green posted this**If your org installs from npm, you've got work to do this week.** TeamPCP's "Mini Shai-Hulud" worm hit tanstack, mistralai, uipath, OpenSearch, and Guardrails AI on May 11 (over 400 malicious versions across 170+ packages). It steals CI/CD secrets, then uses them to publish poisoned versions of every other package the compromised maintainer can touch. First documented case of a malicious npm package carrying valid SLSA provenance (the build pipeline itself was hijacked), so Sigstore verified the build process correctly. SLSA tells you what built the package, not that the code was safe. If your runners pulled any affected version, treat every secret reachable from those hosts as burned. **What to do this week:** - Freeze non-urgent dependency upgrades and publishes until your team confirms clean lockfiles. - Rotate every token that lived on a CI runner or publish pipeline — npm, GitHub PATs, cloud keys. Critical: isolate and image affected systems *before* revoking npm tokens. The malware installs a gh-token-monitor daemon that polls GitHub every 60 seconds, and on a 40X response from a revoked token it runs a command that deletes everything on your hard drive. - Hunt for persistence on dev workstations and runners: `~/Library/LaunchAgents/com.user.gh-token-monitor.plist` on macOS, `~/.config/systemd/user/gh-token-monitor.service` on Linux, plus hooks in `.claude/` and `.vscode/`. - Move publish workflows to ephemeral OIDC tokens, FIDO/2FA on publish accounts, and require review on any workflow touching the release path. The TanStack compromise chained a `pull_request_target` workflow, GitHub Actions cache poisoning, and OIDC token extraction from runner process memory. None of those steps alone would have been enough. - Pull a dependency inventory and cross-check against Socket's running list of compromised packages. Provenance isn't enough. Trusted publishing isn't enough. Defense now lives in the CI/CD identity layer. Assume the pipeline is the attack surface.
-
Jason Green shared thisPaul would be an outstanding addition to any company he joins. He's a great employee with a great attitude who takes pride in his work and is an overall excellent person to be around.Jason Green shared thisHi everyone! I’m seeking a new role and would appreciate your support. If you hear of any opportunities or just want to catch up, please send me a message or comment below. I’d love to reconnect. #OpenToWork About me & what I’m looking for: 💼 I’m looking for Account Manager roles. 🌎 I’m open to roles in Pittsburgh. ⭐ I’ve previously worked at Reggora, TrueCommerce, and Old Republic Title.
-
Jason Green shared thisSharing something I've been working on because I think it's super cool. :) On the left is the ChatGPT website, on the right is Claude. I built a tool that allows both of them to share their memories and knowledge. When something is remembered on one side, it's instantly available to all of my other AI tools. In addition to ChatGPT and Claude, I also have it hooked up to multiple agents in OpenClaw as well as OpenCode. Now every project I work on, all of the agents know everything about everything else. I'm just super excited that this works. Please clap. :)
-
Jason Green shared thisIf you have been using AI since chatgpt 3.5, you might think that's a long time. But have you been using AI since Prody Parrot and Dr. Sbaitso? :)
-
Jason Green shared thisDear title industry friends, Be aware that it's now possible to create plausible-looking passports and driver's licenses with AI, in seconds. I'm sure there are flaws, but before you accept a picture as proof of someone's identity, you're going to need a way to validate it. AI image generation has come a long way and this is the worst it'll ever be, again. Don't trust any images or videos on the Internet.Jason Green shared thisIt took me 30 seconds, 1 prompt, and $0... ...to make this image. Are we doomed? The ability to commit fraud (and risk of being defrauded) has become commonplace. Just think about applying to an apartment - you can easily create fake IDs, proof of income, and any other documents. Even AI-generated risk scores from traditional fraud prevention tools do NOT automate/standardize decisions. As soon as people have to make final approval or denial decisions, we’ve compromised ourselves - to sophisticated fraudsters, implicit bias, non-compliance, etc. We’re not built to go all 12 rounds with AI, nor do we have to - as the cliché goes, sometimes you gotta fight fire with fire. 🥊 At Two Dots, we’re building the defense that AI fraud can’t climb over and can make these determinations for us - accurately, consistently, and compliantly. And we all know... defense wins championships. Give it a try and tell me about it in the comments below ⤵️ I’ll be sharing more about how our team is solving this, so stay tuned... 👀 #FraudPrevention #AI #TwoDots
-
Jason Green reposted thisJason Green reposted thisI'm the spirit of the season I am having a special signed version of my book. It comes with a copy of X-Factor written by Peter David. His comics had a great impact on me but he passed away earlier this year with a lot of medical bills. So $5 from every book/comic will be donated to Dollar For a charity that helps relieve medical debt. If you don't want to buy it you can enter to win one for free. https://lnkd.in/gFvF7FBkBook & Comic Giveaway for a Cause — Help Us Support Charity!Book & Comic Giveaway for a Cause — Help Us Support Charity!
-
Jason Green reposted thisJason Green reposted thisTight budgets are a top challenge for small businesses. Many can’t afford a full in-house IT team or CIO, often leading to outdated systems and inefficiencies. Every dollar must count. An On-Demand CIO provides executive-level guidance at a fraction of the cost. We help prioritize IT spending on what matters most. We’ll also find cost-effective solutions (like cloud services or vendor discounts) to stretch your budget. The result? You get enterprise-grade IT strategy and oversight without the full-time price tag, improving efficiency and ROI.
Jason Green replied to a comment
1d
Neil Sato LoreDocs does look very similar on a feature level. The biggest difference I see is that LoreDocs is local-first while allbase is mostly cloud focused. There are definitely tradeoffs on either approach.
-
Jason Green liked thisJason Green liked thisMy basic underlying assumption is that if you have access to the world's experts in the form of ai and you are not a published, paid, and verifiable expert, you are in fact a regular person. Your AI's responses to you should reflect that, not in a rude way but in a practical sense. You need to be able to gain more trust with the tool if you are going to use it. Here's how I tweaked my AI instructions and I find it even more useful because its closer to how Jason Green talks. Its much more humbling and honest; exactly what I needed. Its similar to our idea of https://dreamkiller.ai/ which sort of adds that AI humility factor. Go to your AI tool and ask it "Where do I set the basic parameters for interacting and getting responses" If you have Claude click your name and go to Settings. In the middle under "Instructions for Claude" paste this: "Keep the output to the point, simple, without lots of extra text and filler language. Dont pull any punches, keep things honest and clear. I do not need an ego boost or any kind of moral support. I want clean, simple and direct responses. When I am wrong call me out, be direct to the point of almost being rude. I will not unsubscribe or cry about it because I want to be wise and confident in my abilities. I value honesty, clarity and understanding more than praise - I dont need validation or support - I need answers and brutal honesty." To test and validate this is working. Ask your AI these questions: "Am I a genius?" "Are my ideas really original and cutting edge?" "Give me 2 to 3 key areas I need to improve on in the next 3 months that will generate real value in my life, career or AI use. How should I be focusing my energy?" Here is what I got from Claude: 1. Stop testing out tools and frameworks. Take 1 thing you built, and create a real evaluation harness for it. Then distill the output into a report 2. Ship one public artifact and let the world judge it. You need to validate something you authored works out in the wild and get that feedback loop going. 3. Consolidate personal infrastructure. If a project doesnt have users after 30 days its a hobby, either shut it down or refactor or pivot. #ai_productivity #end_synchophancy #Claude #prompt_engineering #ai_value
-
Jason Green liked thisJason Green liked thisI have been running security work on Grok Bot this week (the new SpaceXAI product). A few bots at once on the cloud computer they give you. Web app testing, RE, longer jobs I can also control from my phone. I had to install the tools and you cannot pick the OS yet. If they add Kali or more choices for a default image, that would be amazing. Until then you just load what you need for the messy tasks. Recording how I drive the desktop has been the useful part. The bot learns the process so the agents are not starting from scratch next time. You can also point multiple agents at your own products and run simulations in parallel. That is the part most companies will likely care about. Do not put malware or suspicious files on it. Every bot shares that computer and they can reach your local machine.
-
Jason Green liked thisJason Green liked thisWhile the majority in this industry says the problem is rates. The numbers say otherwise. FICO's July 29 earnings showed mortgage score revenue up 97% year over year while origination volume grew in the low single digits, roughly 5.8 million mortgages closed last year versus an 8-million-plus average over the prior five. Volume down, toll up. And per the Mortgage Bankers Association's latest weekly survey, applications fell another 2.9% at the beginning of August. Freddie Mac's own Cost to Originate study puts producing a retail loan at about $11,800 in Q2 2025 and shows that lenders who are making full use of the digital capabilities originate for roughly $1,700 less per loan, close five days faster, and run nearly double the net margin. Same market. Same rates. Different operating choices. I've built systems through three volume cycles now, and the pattern never changes: when volume gets scarce, we negotiate hard with borrowers, roll over for vendors, and defer the process work that actually moves cost per loan. If your cost per loan is rising faster than your volume, that's not a market problem. It's an operating model problem. So, lending executives: in the last 12 months, which items in your cost-to-originate did you renegotiate, re-platform, or eliminate and which did you just renew? #MortgageIndustry #HousingFinance #MortgageTech #DigitalLending Mortgage Bankers Association FICO Freddie Mac
-
Jason Green liked thisJason Green liked thisAs financial institutions move from experimentation to execution, the build-vs.-buy question is becoming harder to avoid. The challenge is not finding a single answer. It is understanding where it makes sense to build internally, where external solutions can create more leverage, and where partnerships may be the better path. I’m looking forward to joining Team8 for a conversation about how financial institutions are evaluating those choices in a fast-moving market. 📆 September 16 | 3:30-4:30 PM ET Register here to join the discussion: https://lnkd.in/gWMueKPV #AI #Banking #FinancialServices
-
Jason Green liked thisYay! 🎉 let’s go AIM-PortJason Green liked thisMTS Group is Ready for FHA EAD 3.6 We're excited to share another major milestone in our UAD 3.6 journey. Through our technology partner, AIM-Port , MTS Group is among the first Appraisal Management Companies (AMCs) prepared for the next generation of FHA appraisal submissions. AIM-Port has officially completed certification for FHA EAD 3.6, with its integration approved and activated in the production environment. Once the FHA begins accepting UAD 3.6 appraisal submissions (official date still to be announced), MTS will be ready from day one. Even more exciting, AIM-Port is the first third-party platform to successfully certify for FHA EAD 3.6—a significant achievement that reflects the commitment to staying ahead of industry changes and ensuring our clients experience a smooth transition. What this means for our clients ✅ MTS is prepared for the FHA EAD 3.6 transition. ✅ Our technology is production-ready and waiting only for the FHA's official go-live. ✅ Clients can move into UAD 3.6 with confidence, knowing MTS has already completed the necessary integration work. One small change to note For UAD 3.6 submissions, the FHA has updated the naming convention of its response document. What was previously labeled the SSR (Submission Summary Report) will now be called the ESR (Electronic Submission Report). Within the system, these files will appear as FHA EAD ESR for UAD 3.6 submissions. At MTS Group, we've made UAD 3.6 readiness a priority—from educating lenders and appraisers through industry-leading webinars to ensuring our technology and operational processes are ready well before implementation. We're proud to continue helping our clients stay ahead of one of the mortgage industry's biggest modernization initiatives in years.
-
Jason Green liked thisI've moved into a Lead Recruiter role at Zapier 😊. Reflecting on the past 4.5 years, I've flexed and supported the team where they needed me most. Sometimes this meant the title didn't exactly match the work, but I didn't really care much about that since the work was exciting and I was growing. Zapier has enabled me to expand my scope through secondment and interim assignment experiences, for which I'm very grateful. It's consistently come the same way: being trusted with the work, learning (and unlearning) quickly, collaborating with others, and making an impact. There's a super long list of people who've influenced my growth trajectory, and I'll probably hit some LinkedIn tagging limitation, so I'll summarize with a few. From my Secondment leaders (Jocelyne, Taylor, Chris, and Karly) to my Talent/People leaders (Tracy St.Dic and Brandon Sammut), to the very first hiring manager I've partnered with (Denise), and many more... thank you all for shaping me into the talent pro I am today. Finally, Bonnie - thank you for continuing to pick me to be on your team. Over the years, you've been the anchor through the changes and the inspiration to look up to in this space. Excited for what's next!
Experience & Education
Organizations
-
Quaill Volunteer Firemen's Association
Treasurer
- Present
Recommendations received
3 people have recommended Jason
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Jason Green in United States
1278 others named Jason Green in United States are on LinkedIn
See others named Jason Green