Jeremy Rule

If you’re still running mission-critical ops on Word docs, spreadsheets, and tribal knowledge… you already know the punchline. Someone skips a step. A part goes missing. A test runs late. Nobody can trace why. And suddenly your “we’ll fix it later” process becomes the bottleneck that stalls a launch, slips a milestone, or triggers an audit you weren’t prepared for. Here’s the part most teams won’t say out loud: Your systems worked when you were 20 people. They quietly collapse when you’re 200. I keep hearing the same frustrations from Heads of Manufacturing, Mission Assurance, Propulsion Test leads, COOs: “We waste hours hunting for the latest procedure.” “Operators keep improvising because our docs are outdated.” “We found an error… but can’t trace where it originated.” “Quality is fine until it suddenly isn’t.” “We think we’re compliant… but we can’t prove it.” And nearly every team believes the same three things: “We can just build our own tool.” Until it takes 10x the time, 5x the cost, and still doesn’t integrate, scale, or get used. “Paper is fine.” But paper is the reason steps get skipped, data gets lost, and audits become nightmares. “We don’t have time to implement something better.” Translation: We’ll keep paying the cost of mistakes instead. Here’s the reality spoken from the front lines: The teams that scale the fastest aren’t the ones with the most people. They’re the ones with standardized, traceable, error-proofed operations. Manufacturing runs on rails. Testing becomes predictable. Operations stop firefighting. New hires ramp in days, not months. And leadership finally sees real-time truth, not guesswork. That’s why teams in space, energy, defense, robotics, and advanced manufacturing are moving to platforms built for complex operations, not duct-taped workflows. And yes, Epsilon3 is one of those platforms that helps them scale. Not because it’s flashy, but because: Setup is fast (2–3 short calls). Imports are automated (no manual rebuild). It works offline and in air-gapped environments. It tracks everything: parts, procedures, schedules, risks, CAPAs. FedRAMP High and enterprise security are non-negotiable. And it replaces a dozen tools you’re already frustrated with. If you’re scaling, hiring, prepping for a government audit, or staring at another avoidable delay… Ask yourself the real question: Can you afford to run next year on the same systems that nearly broke you this year? If the answer is no, let’s talk. I’ll show you how the teams who stopped tolerating operational chaos are now delivering on time, with fewer mistakes, and far less stress. Your future self and your technicians will thank you.

14

As AI continues to reshape how agencies build, deploy and secure software, transparency and deeper risk understanding are becoming essential to getting it right. I recently joined Sonatype Senior Vice President of Product, @Tyler Warden, and Lockheed Martin Vice President and Chief Digital AI Officer, Mike Baylor, for a GovCIO Media & Research #GovFocus virtual panel on the realities of open-source AI — where it accelerates innovation, where it introduces new exposure and what federal teams should be thinking about as they modernize their software supply chains. If you're navigating AI adoption or looking to strengthen your security posture, this discussion offers a grounded look at the challenges and opportunities ahead. Watch “How AI is Reshaping Open Source, Software Supply Chain” here: https://lnkd.in/es9xv9Wi #SupplyChain #FederalIT #GovTech #AI #ArtificialIntelligence

109

1 Comment

An unspoken truth about ensuring software just "works" -- engineers default to making infrastructure as rigid as possible. But in an era where we're building faster than ever, that mindset needs to be challenged. Capture knowledge while you build. Then use it to make the team and the product better every single day. I'm proud of our team and what we are doing. P.S. Anthropic should maybe check out OpsCompanion with their recent troubles 😉

19

2 Comments

A pilot side-hustle boom is teaching us a massive lesson in skills adjacency. When we look at the data, a pilot’s second job isn't random. They cluster in: ➡️ Real Estate (Top side hustle at 6.7%) ➡️ Entrepreneurship ➡️ Education/Training ➡️ High-stakes Sales The insight: These professionals aren't just flying planes; they are experts at monetizing trust, schedule autonomy, and relationship management. The missed opportunity: If your internal mobility team isn't building pathways for these folks into Sales Engineering, Operations Leadership, or Corporate Training, they will sell those capabilities elsewhere. Stop looking at job titles and start looking at adjacent capabilities. Which roles in your business are you leaving unbuilt? https://lnkd.in/eMrAtQDV #FutureOfWork #OrgDesign #InternalMobility #WorkforceData

4

Seven weeks "retired." My output since then disagrees with the title. Three production apps shipped. DIAL (DORA metrics + AI recommendations), a church leadership assessment platform, and an internal time tracker — all built solo, all built right. My non-negotiables across every project: ✓ Firebase + GCP — managed infra, zero ops overhead ✓ Workload Identity Federation — no stored keys in CI/CD ✓ Angular 17 standalone components + consistent design system ✓ Loosely coupled services with clean API boundaries ✓ Conventional commits enforced via Husky Modern tooling lets one engineer build like a team. #Firebase #GCP #Angular #SoftwareArchitecture #FractionalCTO

71

2 Comments

Palantir is hiring 18-year-olds straight out of high school. No college degree required. No "4 years of prerequisites." Just coding skills and "diversity of thought." $5,400/month to skip university entirely and start building software for defense tech. Palantir CEO Alex Karp: "Every single system is parasitic. Our job is to break that." Bold? Absolutely. Controversial? Definitely. Worth paying attention to? 100%. Here's what's interesting: 500+ high school seniors applied to their first cohort. 22 were accepted. "More than a handful" are getting full-time offers. And according to their head of talent, many of these 18-year-olds have coding skills BETTER than Palantir's college graduate hires. Karp's message to universities: "If we scale this to 80-90 people, that's a real problem for the universities." At MyTrencher, we're watching this closely. Not because we think universities are obsolete. But because it highlights a fundamental question: - What's the actual value proposition of a 4-year degree in 2025? For some students: → Critical thinking, networks, research opportunities = invaluable → Liberal arts education = foundation for life For others: → $100K+ in debt for skills they could learn elsewhere → 4 years when they're ready to build NOW The reality? Both can be true. The issue isn't "college vs. no college." It's that the traditional path has become: - Too expensive - Too slow - Too disconnected from what companies actually need Palantir is betting they can find exceptional 18-year-olds and train them faster. Maybe they're right for SOME students. Maybe this works for SOME companies. But here's what we know for sure: The old model is breaking. And companies like Palantir are forcing universities to answer hard questions. What do you think? Bold innovation or dangerous shortcut? DROP a comment below...

9

2 Comments

You don't have to be trained as a software engineer to build cool things anymore. Bring a curious mind, a problem solving mentality, and a grasp of the English language, (here in the US, anyway!), and you can do amazing things...

3

#DeliveryExcellence The loudest celebration in most programs happens on launch day. The quietest failure often happens in the weeks that follow. Go-live is a milestone, not an outcome. I have been part of programs where the deployment was flawless, the release notes were clean, and the client signed off on every acceptance criterion. Two months later, adoption was at 30 percent and the business was running parallel processes on spreadsheets. The gap between a successful deployment and a successful program lives in what happens after launch. Hypercare is where you discover that the training was not enough for edge cases. Stabilization is where you learn that performance under real load looks nothing like performance in a test environment. Adoption tracking is where you realize that "available" and "used" are very different words. In my experience, the programs that sustain their impact beyond go-live are the ones that plan for this phase with the same rigor as the build. They allocate dedicated resources for post-launch support. They define adoption metrics before release, not after. They treat the first 60 days in production as part of the delivery scope, not as an afterthought. When I set up delivery governance for a new program, I always ask one question during planning: what does success look like 90 days after go-live? If the answer only references deployment, we have a gap. The launch is not the finish line. It is where the real feedback begins. #GoLive #DigitalTransformation #ProgramManagement

13

I strongly suspect all the various vendor-promoted posts that "we don't need bigger LLM contexts!" are complete and utter crap. Or if not outright crap, don't apply to all models. I don't think it accounts for multi-head attention or MoE, or in some models, both. A large context isn't about shoving more junk through the same pipe, it is about allowing segments of what is in that context to matter differently to different focuses of attention. LLMs may not be the path to AGI, but I don't think it helps to get game-of-telephone'd perspectives that were fine when the only thing you wanted an LLM to do was be a simple chat UI query engine dealing with short but highly repetitive operations. For those, a model mostly just needed to be a good human language mimic engine. Small contexts are great for "what is the weather forecast for today?" If you want to do hard problems well, like actual research problems (and maybe a lot of software engineering problems) I'd argue that what is needed is more like a context 100 times as big. Go entirely the opposite direction the industry has been focused on. Because with a multi-head/MoE mix like you have with DeepSeek, that's going to be the closest I think you'll get to how people consider non-trivial problems from multiple perspectives and then integrate the various views after. It's hard to simulate anything even marginally close to a brain with multiple areas of specialization, when the simulation doesn't have the elbow-room to allow those specializations to have full effect. Instead of figuring out where to find another trillion tokens to scrunch into the same parameters, maybe just figure out what would let the existing data do more. Which, if you think about it, prior to LLMs that is how we normally treat data. As a pool of untapped value that we should do more with. Instead we're in this situation where industry is acting like "building bigger buckets yields more value!" No, doing more with what is *IN* the buckets yields more value.

1

Some of the best advice I ever got from seasoned OSS leaders was to stop doing work that others on the project could do instead. Step back, let others step up and focus my energy on higher impact problems. This advice proved even more applicable as CTPO and cofounder of Workbrew, too.

21