Bjørn Bjerkøe

Proven Methodology from Four Decades of Experience Get my book, "Software Development Playbook" for FREE! This book is NOT ABOUT PROGRAMMING. Drawing on nearly 40 years of professional software development, the book distills a repeatable, battle-tested formula for planning, executing, and delivering projects. This framework helps you avoid common pitfalls and ensures your team follows best practices from day one. https://lnkd.in/ey_9-XAF #softwaredevelopment #projectmanagement #markturkel #palmbeachsoftware

5

I spend too much time orchestrating coding agents that write, review, and fix code. Generally, just copying and pasting from one agent to another. Creating a local app to streamline the entire SDLC pipeline. Output from one agent is the input to the next agent until a PR is complete. For example, once a PR is created, my CI/CD pipeline runs the code review process using Codex, Gemini, Claude code, and SonarCube. This local app would review the comments and SonarQube output, decide which feedback to incorporate, and resolve the comments. It replies to the comment before resolving, creating an audit trail. If a review item is not safely fixable, the agent should leave an explanatory GitHub reply instead of silently skipping it. I'd envision this app will eventually orchestrate the entire development pipeline. Orchestrating the entire pipeline from 1. Project management tool 2. Brainstorm with the user 3. Code 4. Lint 5. Test 6. Dogfooding - let the agent try the app 7. Deploy to preview / staging 8. Run PR review process with multiple review agents 9. Get feedback from the team using preview links 10. Generate demo videos, changelogs, and help center articles 11. Deploying to production safely 12. Read logs and screen recordings to fix bugs or UI issues while you are asleep 13. Keep a record of lessons so that we don't make the same mistakes twice link in comment

8

1 Comment

🔎 BRD, FRD, SRS, TDD… why do these terms trip so many people up? As Business Analysts, our role isn’t to “fill templates.” It’s to turn business goals into something the tech team can actually deliver. 💡 Here’s how I break it down: BRD (Business Requirement Document) → The goalpost: What problem are we solving for the business? FRD (Functional Requirement Document) → The behavior: How should the system act to meet those needs? SRS (System Requirement Specification) → The detail: Rules, conditions, and exact scenarios the system must handle. TDD (Technical Design Document) → The construction plan: The step-by-step design for developers to build it. 👉 When these flow in the right order (BRD → FRD → SRS → TDD), you cut confusion, avoid endless rework, and keep everyone—from business to developers—on the same page. 🚀 My takeaway: A strong BA doesn’t just produce documents. We connect strategy with delivery, ideas with action, and vision with trust. #businessanalyst #fintech #mutualfunds #goldtech #wealthmanagement #requirements #digitalfinance #businesstransformation #careergrowth

97

4 Comments

They were whispering the answers off camera?!?! Let me explain... When selecting a software development sub-contractor or "development partner", how do you really know if they're any good? When hiring a developer as an employee, there will usually be a hands on element of the interview. Using this approach I've built exceptional in-house teams for clients, with resulting teams that benchmark at the top of the scale, let alone top of the industry (with the DORA Software Delivery Performance '4 keys' metrics – see: https://lnkd.in/eSf8zeAW ). Do you do the same with your suppliers/partners? If you don't, the resulting technical debt you inherit might not be apparent until long after the last of their invoices are paid and your in-house teams take over (seen this so many times)! Years ago, a client asked us to do such an evaluation. The supplier (a big name) was asked by our client to arrange for us to pair with the people that the supplier had in mind for the engagement. The results were shocking... Someone could be heard whispering instructions in the background! Even those instructions fell short of even basic competency. The consultants they had in mind were barely "advanced beginners" and certainly fell short of "competent" (see the Dreyfus Model)... And this was despite the supplier's sales people having talked a good talk, claiming they had the best people in the business. Suffice to say, the client went elsewhere. That was a smaller engagement... For larger engagements, I would start with a DORA 4 keys evaluation of a sample of the supplier's teams, which I'd also ask to see evidenced. Then, I'd select a random sample of those teams to mob-program with to reinforce that evidence. Only then could I endorse that supplier... So, when selecting your subcontractor/partner, just keep in mind that the firms with the best sales people with the biggest budgets to wine and dine you sometimes have the worst consultants... And sometimes the companies with the worst sales people, who sell on their people's merits and not expensive meals, have the best consultants. Navigating perceptions vs reality is the key. As they say, never judge a book by it's cover.

1

The AI productivity conversation has me wondering about a potential 'Big Uh-Oh' hiding in plain sight. Not taking sides here... just surfacing a tension I've seen in debates across QA, dev, and exec teams (customers as well as groups here on LinkedIn). Curious what y'all think (I know, a New Englander using "y'all"... cut me some slack)? The Root Tension: Production just moved to machine speed. Verification is still running at human speed. That's not acceleration... that's a high-velocity bottleneck. Your AI is generating, say, maybe 10x the output. But every unit still needs a human to determine if it's trustworthy. The review queue doesn't shrink because the tool is faster. It explodes. AI is the accelerator... a reinforcing loop driving volume up fast. Human verification is the balancing loop... the system's natural stabilizer. When those two forces fall out of balance, the system doesn't just slow down. It fails. The "Teenage Son" Analogy: My colleague's sage wisdom about his son comes to mind: "I trust him & I raised him right. But he's a teenage boy... they're inherently untrustworthy. So I love him, I trust him... and I verify the big things. Not because he's bad. Because he's unproven." To me it's a near perfect analogy. AI is exactly the same... well ... for now. Both will mature over time, but unproven systems require discipline, not just hope. QA Discipline: In my QA life, I run regression tests even when I expect zero bugs. You change the system, you validate the system. That's not distrust. That's discipline. The Balancing Loop Challenge: Corporate response right now seems to be cutting their "verification people" indiscriminately. Removing the "Human Hardware" exactly when production load is spiking. That's not tuning the balancing loop. That's disabling it. and before you say "just use a second AI to check the first one"... that's not balancing. That's two accelerators agreeing with each other. One machine validating another's "plausible enough sounding" output isn't verification. It's an echo chamber. AI can tell you if the code is well-formed. It can't tell you if it's right for the business. The machine knows the prompt; the human knows the consequences of the prompt being wrong. My Questions: Is the goal of sustainable AI productivity with certainty mean fewer humans or more? Lowering unit verification cost? Better craft? Smarter tooling? Architectural constraints that reduce what needs reviewing? Is anyone actually investing in the balancing loop (and How?)... or just riding the accelerator hoping nothing breaks? I am honestly seeking clarity & discussion here. I'm not against the accelerator. I just want to be cautious about a car with no brakes (every LinkedIn post needs a cliché, right?).

5

In program reviews, engineers always hit technical debt first because outdated interfaces paired with crusty legacy code create headaches when bolting on new tools. But something deeper holds everybody back. Let's call it innovation debt. You see it when leaders sometimes fail to help engineers adapt to their systems, and the outfit can slow to a crawl. That is why those same legacy systems take fire in briefings, even though they have decades of battle-tested logic baked in. Engineers hammered them out under insane deadline pressure, where one bug could mean lives lost. No wonder teams rush to swap them out and lose all that tribal knowledge in the process. Programs typically crater from knotted-up ops, regs, and burned-out folks carrying gaps nobody wrote down. Look at the fallout, and you see complexity running too many programs. True change sticks when leaders respect engineering systems and culture by training them for tomorrow's challenges, even though we still tend to measure technical debt mainly by lines of code. Meanwhile, innovation debt usually shows up as burned-out engineers and sluggish processes that can break trust. The actual fix means peeling back those layers instead of chasing shiny quick fixes.

15

If your automations keep breaking, it’s not you — it’s your stack. I see this all the time. People spend hours (sometimes days) trying to make automation tools do what they want. Scripts break. Workflows fail. Frustration grows. One friend recently told me he ran the same automation 38 times before it finally worked. Thirty-eight. The problem wasn’t the code. It was the tool. He used N8N, a powerful platform, but for his goal, Make.com would’ve done it in half the time. No API headaches. No missing nodes. Just results. That’s the thing about automation: 👉 The most advanced tool isn’t always the right one. 👉 Time is a bigger cost than features. 👉 Progress matters more than perfection. When it comes to automation, start simple. Prove the value. Then grow complexity only when it earns its place. Because speed to value will always beat sophistication without success. If you’re spending more time fixing automations than benefiting from them — your tech isn’t helping you. It’s managing you. Audit your stack this week. Ask yourself one question: “What’s the simplest tool that can do the job right now?” You’ll be amazed how much time you’ll get back. PS: If you want to go deeper into advanced automation, we’re hosting a two-session live course on December 1 and 8 at 12 PM ET — two hours each. Join us to learn how to build faster, smarter, and more reliable automations. Link in the comments for full details.

13

13 Comments

https://lnkd.in/gDG_yQga I mentioned to several ex-coworkers a year ago that really good QA folks are going to be great at Specification Driven Development (SDD) as they can be the layer between the daydreamers and the end product. But, if you are a good developer and a great daydreamer, you can daydream reality if you implement SDD methodology.

2

Every process issue I solved had one root cause: People. AI isn’t the blocker. Tech stacks aren’t the reason initiatives fail. The real breakdown happens much earlier. It starts with misalignment. When teams aren’t speaking openly. When clarity gets replaced by speed. When leaders avoid difficult conversations. Because the process isn’t usually the problem. It’s what’s happening around it. Most often, it’s: ✓ Misaligned teams ✓ Poor communication ✓ Leadership anxiety disguised as over-analysis Fix the people, and the process usually works. AI adoption is no different. If your people are unsure, unprepared, or unaligned, no algorithm will save you. Repost ♻️ to help someone catch the real problem. P.S. Do you check people-fit before process-fit?

37

45 Comments

If I could add one to the list: "Migrating off the software system we've been using for 20 years" There's a better way: maintain what you've got and save the headache of a big software migration! #legacysoftwaremaintenance #softwaremigration #fractionalcfo

7

6 Comments

Nate's one unique trick for leaders to leverage AI One opportunity that feels missed today is the tremendous availability of quality engineers in between positions right now. I have a number of former peers and friends that can be solid contributors to your codebase. AI makes the onboarding process so much easier than it has ever been. Look through your team's backlog, find a feature or service that doesn't touch any PII, doesn't add too much operational risk to your team. Give a candidate a two week contract. See what they can do. Maybe it doesn't work, but I bet there's a good chance you move the ball forward for your company. We worry so much about false positives in the hiring process, but we never test that theory. This could be a low stakes way to test that assumption. #oneteamoneheart

8

New Cypress blog post showing how to mock the locked down "location.href = ..." assignments in your E2E tests: "Mock The Location Href Property" at https://lnkd.in/eRjSYNfn

11