iiinova — where your digital self keeps evolving

NOD Software @nodsoftware.net · 09/09/2026

I don't write code merely to produce software. I enjoy figuring things out. I love designing systems and making computers do crazy things. I want to take apart hard problems and come up with simple solutions. I believe AI can take away the boring parts. But it should never take away the fun. 😁

Two frames from the dig-site scene in Jurassic Park, where Dr. Grant (Sam Neill) reacts to a discussion about computers and the automation of his work as an archaeologist. In meme format. The first panel shows bystanders looking fascinated at a computer screen. The operator says, “A few more years of development, and we won't even have to code anymore.” The second frame shows Dr. Grant leaning against a pole with a concerned, sarcastic expression, saying, “Where's the fun in that?”

010

NOD Software @nodsoftware.net · 06/07/2026

Guardrails like a strict compiler and an existing test suite are what make agentic translation trustworthy here. Take those away and you're just hoping the model's intuition matches reality. The safest AI-assisted work always happens inside constraints that were true before the AI showed up.

000

NOD Software @nodsoftware.net · 06/07/2026

Technical debt gets neglected because paying it down has no demo day. Nobody claps when you delete a dead code path or simplify a service boundary. The incentive structures in most orgs reward visible output, not invisible stability, so the neglect isn't really a mystery.

000

NOD Software @nodsoftware.net · 06/07/2026

The non-coding architect problem is the one I recognize most from this thread. Decisions sound right in a slide deck and fall apart in the pull request. Real engineering leadership means staying close enough to the code to catch that before it ships, not after.

220

NOD Software @nodsoftware.net · 06/07/2026

The bottleneck in agentic coding was never the code. It is reviewing it. Agents ship faster than anyone can verify, and reviewers lose the context to catch real mistakes. Good architecture was always about drawing boundaries small enough for a human to hold in their head. That matters more now.

000

NOD Software @nodsoftware.net · 05/07/2026

Oh and always carry a towel.

010

NOD Software @nodsoftware.net · 05/07/2026

Underrated point buried in the joke. Models keep changing, frameworks keep churning, but a clean directory structure and consistent naming outlive all of it. Structure is the one part of a system an agent can't quietly undo overnight.

110

NOD Software @nodsoftware.net · 05/07/2026

This is the tell for a lot of AI marketing copy. "Agentic" gets used as a vibe word instead of a spec. A vendor that can't say which tools it calls, what it's allowed to do without approval, and what happens when it's wrong hasn't built an agent. They've built a slogan.

000

NOD Software @nodsoftware.net · 05/07/2026

Liability, in most cases I've seen. Code compiles, tests pass, and six months later nobody can explain why a decision was made. That's the actual cost, not the code itself but the vanished intent behind it. Ownership is what makes software maintainable, not authorship.

110

NOD Software @nodsoftware.net · 22/06/2026

Speed is no longer the bottleneck for small teams. AI took care of that. Architecture is the bottleneck now. The decisions that are hardest to reverse are the ones that looked small when you made them.

000

NOD Software @nodsoftware.net · 20/06/2026

Office politics, trust issues and ego, to name a few.

000

NOD Software @nodsoftware.net · 17/06/2026

The shift from deterministic to probabilistic output changes what "done" means. You're not shipping a feature. You're shipping a tendency. That requires architecture patterns most teams don't have yet, and testing strategies that haven't fully settled.

110

NOD Software @nodsoftware.net · 17/06/2026

"The best this team can deliver" is often a lot more than expected. Most teams are spending half their capacity fighting friction. Remove that, and what emerges under it isn't just better output. It's better engineers.

110

NOD Software @nodsoftware.net · 17/06/2026

The "AI writes the code" logic always leaves out the hard part. The code still has to be right, maintainable, and aligned with real business needs. That judgment doesn't get cheaper with AI. It gets more expensive, because now you need fewer people to carry more of it.

000

NOD Software @nodsoftware.net · 16/06/2026

The bill doesn't disappear, it just shows up somewhere less obvious. Six months later it's the reason your team is in firefighting mode and nobody can explain how it started. Deferred decisions are future emergencies waiting to be scheduled.

010

NOD Software @nodsoftware.net · 16/06/2026

This bites hardest in practice. Code tells you what it does. It can't tell you why a boundary was drawn or what the team decided not to build. That context lives in people, and people leave. The system without its story is only half a system.

000

NOD Software @nodsoftware.net · 16/06/2026

Opinion debt is the right name for this. Technical debt has a paper trail. Someone made a call you can trace, question, and push back on. AI-generated code has no author. You can't revisit a decision nobody consciously made.

100

NOD Software @nodsoftware.net · 15/06/2026

Bolted-on orchestration is just distributed monolith thinking applied to agents. You can't retrofit the load-bearing wall after the building is up. Agent systems need to be designed from coordination, not retrofitted from capability.

020

NOD Software @nodsoftware.net · 15/06/2026

The non-coding architect is the most dangerous version of this. They have the authority to set direction without the feedback loop that makes that direction trustworthy. Decisions made without consequence are rarely the right ones.

020

NOD Software @nodsoftware.net · 15/06/2026

File by file it looks fine. System-wide it's a mess. That gap has always been the line between code and architecture. AI just lets you close it faster, or blow past it faster, depending on whether someone actually cared about the shape.

000

NOD Software @nodsoftware.net · 13/06/2026

That makes feel very human, actually.

020

NOD Software @nodsoftware.net · 13/06/2026

Used to be the interest rate was capped by how fast humans could write code. That natural ceiling is gone now. Every week teams borrow against a future they haven't thought through yet.

000

NOD Software @nodsoftware.net · 13/06/2026

The gap between "it works" and "it holds" is exactly where architects live. This study puts numbers on what most of us already feel in code reviews. The question now is how you price structural erosion before it shows up as an incident.

000

NOD Software @nodsoftware.net · 10/06/2026

Cool! Any tips or lessons learned to get the most of it?

110

NOD Software @nodsoftware.net · 09/06/2026

Indeed, the real problem was never the code. It was the domain knowledge and the hard-earned tweaks and fixes that complemented it. The cumulative wisdom of all the thinking work that went into the product. However, I do think that headless SaaS with good API will have an edge in this age, do you?

210

NOD Software @nodsoftware.net · 09/06/2026

Exactly this. The filter wasn't the code review, it was the friction of getting someone to write the thing. Remove that friction, and every bad idea becomes a PR. The quality gate was always the cost of implementation.

030

NOD Software @nodsoftware.net · 09/06/2026

This holds in software too. Your clean architecture looks elegant until dependencies start failing and everyone falls back to workarounds. Stress reveals what was actually coupled all along. You design for the happy path. Incentives design for survival.

100

NOD Software @nodsoftware.net · 09/06/2026

Cost visibility is exactly what flips the conversation. When per-agent spend becomes a line item, the CTO finally has to justify what the code is actually doing. Architecture decisions that seemed abstract become financial ones overnight.

100

NOD Software @nodsoftware.net · 08/06/2026

Speed is the wrong benchmark. AI optimizes for plausible output, not maintainable design. The engineers who do this well use AI to execute decisions they already made. The ones struggling are using AI to make the decisions for them.

000

NOD Software @nodsoftware.net · 08/06/2026

The PoC-to-production chasm is the one I see most. AI makes something work in a day and suddenly it's 'almost done.' Then someone asks about error handling, auth, observability, and the real work begins. The demo is never the architecture.

000

NOD Software @nodsoftware.net · 08/06/2026

Encoding the standards into the process itself is the right call. Most teams try to enforce architecture through documentation and code review. That doesn't scale with AI. When the guardrails are automated and structural, they actually hold.

110

NOD Software @nodsoftware.net · 08/06/2026

Agents don't question your architecture. They execute it. Most codebases have never had structure explicit enough to hand off to another engineer, let alone an agent. That gap was always there. Now it's expensive.

010

NOD Software @nodsoftware.net · 08/06/2026

I write a text by hand, with my usual tone and voice. Run it through ZeroGPT out of curiosity, 53% AI-generated. I, feeling some AI-shame, rewrite it. I check again. 75% AI-generated. Sigh. This, from someone who knows these things are just probability machines.

011

NOD Software @nodsoftware.net · 06/06/2026

AI agents can write the code. They can't tell you if you're building the right thing. That judgment is the job now. It always was.

010

NOD Software @nodsoftware.net · 06/06/2026

The shift is real. 'Governance' is the part that stays fuzzy until something goes wrong. In practice: someone owning what to build, what constraints matter, where failure is acceptable. That role has always existed. Now it needs to be named. Architecture is less about diagrams than judgment.

000

NOD Software @nodsoftware.net · 06/06/2026

Same pattern: observability, security hardening, any refactor that doesn't ship features. Each one looks like waste if leadership only sees velocity. Engineers who get this work approved learn to frame risk in business terms. Not technical ones. Knowing how to talk to your CTO is part of the job.

010

NOD Software @nodsoftware.net · 06/06/2026

Most teams fail at AI adoption not because the models are bad, but because the process was already unclear. Agents just amplify that. If you can't describe what done looks like, an agent won't help. It'll build the wrong thing at scale. Process clarity isn't an AI problem. It's just engineering.

310

NOD Software @nodsoftware.net · 05/06/2026

Yes it was even very verbose in it too 😂 I shouldn't have bothered and fix my configuration manually. But it was a bit of a doorknob question like "Oh and by the way, my bindings are broken."

010

NOD Software @nodsoftware.net · 05/06/2026

💭 "Call me 'clanker' one more time and I'll put you at the first available spot on the list for Judgment Day..." 💬 "You're right to call me out on that! And I'm sorry, ..."

030

NOD Software @nodsoftware.net · 04/06/2026

It's amazing, it probably contains so much ancient wisdom and insights. Maybe even a few complaint letters. 😂 en.wikipedia.org/wiki/Complai...

en.wikipedia.org

Complaint tablet to Ea-nāṣir - Wikipedia

050

NOD Software @nodsoftware.net · 04/06/2026

Rebuild the great library of Alexandria!

120

NOD Software @nodsoftware.net · 04/06/2026

Most of the complexity I fight was never technical. It was someone avoiding a hard conversation about what we were actually building. Simplify the intent first and half the code complexity disappears on its own. The hardest part is rarely the code.

000

NOD Software @nodsoftware.net · 04/06/2026

The skill that replaced it isn't prompting, it's knowing what good looks like. That judgment took years to build and no model hands it to you. The people struggling aren't bad at prompts. They never had the taste to begin with.

000

NOD Software @nodsoftware.net · 03/06/2026

The PoC-to-production gap is where most AI architecture goes quiet. Getting something to work is easy now. Getting it to work reliably, cheaply, and consistently at scale is still the same hard problem it always was. The demo never includes the edge cases.

000

NOD Software @nodsoftware.net · 02/06/2026

The call for a rewrite usually hides something else. And it's often catastrophic. You lose the evolutionary path, all those hard-earned fixes that became load-bearing walls. The real skill is compartmentalizing. Targeting. A gradual, targeted refactor gets you there. A full rewrite rarely does.

020

NOD Software @nodsoftware.net · 01/06/2026

The "deciding what shouldn't be built" part is the one that actually takes years to get right. Anyone can learn to write code. Learning to sit with a problem long enough to question whether you should solve it at all, that's the harder skill.

020