iiinova — where your digital self keeps evolving

Andy @amonimous.bsky.social · 06/10/2026

Architecture. This work is side-project based. I am focused on building iteratively. If it becomes something, that’s where I feel scaling will help.

110

Andy @amonimous.bsky.social · 06/10/2026

It’s not as fast as fully automated approaches, but it has a couple of big advantages for me: 🔹 Token friendly 🔹 Lightweight 🔹 Small, achievable steps 🔹 I stay involved in the important decisions But I’m starting to think the specific framework matters less than I originally thought.

000

Andy @amonimous.bsky.social · 06/10/2026

The important thing to do is isolate and protect yourself from change. Change in LLM, change in agents, change in harnesses and change in frameworks. The process is what matters. The project is only as good as the documented design.

000

Andy @amonimous.bsky.social · 06/10/2026

For me, GSD has been a particularly good fit as a solopreneur. It gives me enough structure to keep moving forward quickly without introducing a huge amount of overhead. I can stay vendor agnostic, scale if this project becomes successful and adopt to different workflows as the tech evolves.

000

Andy @amonimous.bsky.social · 29/09/2026

I’ve spent a lot of time experimenting with different AI development methodologies, trying to find something that actually works for my projects. For me atm, the combination of an AI Agent + playbooks is producing the best results.

000

Andy @amonimous.bsky.social · 29/09/2026

Clear acceptance tests, easy to set up and maintain, no agent-to-agent coordination, small etc However, I now spend far more time thinking about what should be builtA and that reminds me of the difference between a great photographer and everybody with a camera phone.

000

Andy @amonimous.bsky.social · 29/09/2026

Everyone can take a photo now. The scarce skill is knowing what to photograph, when to take it, how to frame it, and what you’re trying to say. AI has made writing code increasingly accessible. The interesting engineering skill is becoming knowing what code should exist in the first place.

000

Andy @amonimous.bsky.social · 29/09/2026

I haven’t hand-written code since May 25th. That still feels slightly strange to say. But getting here hasn’t been straightforward #buildinpublic #ai #aidevelopment #softwareengineering #vibecoding #llm #coding #developerworkflow #aiprogramming #agents #futureofcoding #buildinpublicjourney

430

Andy @amonimous.bsky.social · 17/09/2026

It’s the amount of context required to make a large solution work. * Too much up-front context You need to explain an enormous amount before the AI can make sensible architectural decisions.

000

Andy @amonimous.bsky.social · 17/09/2026

* Too much to understand afterwards Even if it produces something that works, there’s a significant amount of code, structure and decisions to understand and validate. I’m increasingly favouring, simply guided build through incremental changes, expressed as small stories. Think Gherkin-ish

000

Andy @amonimous.bsky.social · 17/09/2026

Small changes that can be implemented, tested and understood independently. Both the AI and I can follow the stories and the results, meaning there is less to synchronise afterwards. The current tooling simply isn’t at the point where large-scale system design can reliably be delegated in one shot.

000

Andy @amonimous.bsky.social · 17/09/2026

One-shotting a proper solution with AI hasn’t worked for me. The problem isn’t necessarily the quality of the model. #buildinpublic #ai #aidevelopment #softwareengineering #vibecoding #llm #coding #developerworkflow #specdrivendevelopment #systemdesign #futureofcoding #buildinpublicjourney

530

Andy @amonimous.bsky.social · 09/09/2026

And nail the 80% consistently

000

Andy @amonimous.bsky.social · 08/09/2026

But honestly? Sometimes it feels a bit like smoke and mirrors. In my own development, AI absolutely boosts my productivity.

000

Andy @amonimous.bsky.social · 08/09/2026

1x → 2x → 5x? But I keep falling into the same trap… I give AI too much responsibility. The project grows. The AI makes more assumptions. The architecture starts drifting. And suddenly I’m in vibe hell, trying to untangle thousands of lines of code that I didn’t really intend to create.

000

Andy @amonimous.bsky.social · 08/09/2026

The frustrating part is that AI is so good at producing code that it’s tempting to let it run. But fast code isn’t necessarily fast development. Sometimes the fastest route is Stop. Think. I’m still trying to find that sweet spot between AI leverage and AI overreach.

000

Andy @amonimous.bsky.social · 27/08/2026

The idea of shifting left makes sense. In theory, this should save time later. But there’s a problem… Specification fatigue. The more time spent describing exactly what needs to be built, the more attention and energy it consumes

000

Andy @amonimous.bsky.social · 27/08/2026

And some LLMs overthink the problem with endless questions. Eventually, quality can start to drop. And honestly? Coding is far more engaging than writing specs. So perhaps we’re still using the wrong tools for the job. What if requirements and specifications could become more like coding?

000

Andy @amonimous.bsky.social · 27/08/2026

Write the requirement. Press ▶️. See the result. Debug it in real time. Maybe the future isn’t just AI writing code. Maybe AI helps us build executable requirements. That could make specification development more interactive, more engaging — and ultimately, more useful.

000

Andy @amonimous.bsky.social · 27/08/2026

A challenge I’m starting to feel with AI-assisted development: Coding is far more engaging than writing specs. #buildinpublic #ai #softwareengineering #specdrivendevelopment #requirements #developerworkflow #llm #systemdesign #futureofcoding #aidevelopment #coding

340

Andy @amonimous.bsky.social · 18/08/2026

This is worth remembering. Good software engineering hasn’t disappeared. If anything, AI makes some of the fundamentals even more important. Decomposition Break systems into: • Small, understandable components

000

Andy @amonimous.bsky.social · 18/08/2026

• Independently testable parts • Independently deployable changes Then keep testing continuously. AI can be particularly useful here — not just writing tests, but actively helping to model threats, challenge assumptions and look for vulnerabilities. Build → Test → Security checks → Deploy

010

Andy @amonimous.bsky.social · 18/08/2026

Small changes. Fast feedback. Continuous validation. My target? From commit to production in under 10 minutes. The faster you can safely validate and deploy a small change, the faster you can respond. architecture, testing, security, observability and disciplined delivery still matter.

000

Andy @amonimous.bsky.social · 18/08/2026

A recent discussion around AI, open source and security. The challenge, how to keep up when AI find vulnerabilities continuously. #buildinpublic #ai #softwareengineering #cybersecurity #appsecurity #devsecops #cicd #opensource #softwarearchitecture #llm #aidevelopment #continuousdelivery #devops

460

Andy @amonimous.bsky.social · 17/08/2026

We’ve moved from: Machine code → Assembly → 3GLs → 4GLs Each step takes the developer further away from the raw machine and closer to expressing what we want. However, AI can also: • Extrapolate from incomplete requirements • Infer intent • Generate code ahead of where you’ve explicitly specified

000

Andy @amonimous.bsky.social · 17/08/2026

• Explore implementation options • Reason about the solution That’s certainly more than simply another abstraction layer. It feels closer to a 5GL. But there’s an interesting limitation today: The human is still very much in the loop. We provide the intent. We review the output.

000

Andy @amonimous.bsky.social · 17/08/2026

We correct the AI. We decide whether the result is acceptable. But what happens as the models improve? Could AI eventually move beyond generating source code at all? Perhaps the future looks more like: Human intent → AI reasoning → Intermediate representation / bytecode → Execution

000

Andy @amonimous.bsky.social · 17/08/2026

Where the traditional programming language becomes an implementation detail rather than the primary interface. Maybe AI-native development isn’t the next programming language. Maybe it’s the beginning of programming without languages.

000

Andy @amonimous.bsky.social · 17/08/2026

Is AI-native development just the next generation of programming language? Development languages have steadily become more abstract over time. #buildinpublic #ai #softwareengineering #programming #programminglanguages #llm #aidevelopment #futureofcoding #developerlife #technology #genai #5gl

430

Andy @amonimous.bsky.social · 13/08/2026

For years, I’ve struggled with projects that were slow to get started. You have the idea. You know what you want to build. But getting from idea to something meaningful takes time.

000

Andy @amonimous.bsky.social · 13/08/2026

AI has completely changed that. Now I can go from an idea to a working system incredibly quickly. And yet… When everything is built for me almost instantly, I sometimes feel detached from the result.

000

Andy @amonimous.bsky.social · 13/08/2026

There’s less of the struggle. Less discovery through implementation. Less satisfaction from finally getting something working after wrestling with it. So I’m asking myself: Is this AI apathy? Or am I simply using AI the wrong way? Perhaps the goal isn’t to have AI do everything.

000

Andy @amonimous.bsky.social · 13/08/2026

I’ve started wondering if there’s a new problem with AI-assisted development: AI apathy. I find myself conflicted sometimes. #buildinpublic #ai #softwareengineering #coding #developerlife #llm #aidevelopment #programming #futureofwork #devthoughts #productivity

450

Andy @amonimous.bsky.social · 21/07/2026

Who said my code wasn’t beautifully architected ;-)

000

Andy @amonimous.bsky.social · 21/07/2026

AI takes that idea a step further. Instead of just providing building blocks, it can generate large parts of the implementation itself.

000

Andy @amonimous.bsky.social · 21/07/2026

The result? On my personal projects, I’m making far more progress than I used to. But there’s an interesting trade-off. I sometimes find less joy in the process. Not because the results are worse. But because I enjoy the craft of building, understanding, and solving problems myself.

000

Andy @amonimous.bsky.social · 21/07/2026

So I’ve been wondering… Is that simply my own desire to code? Or is there something important about being deeply involved in creating a solution that AI can’t replace? I’m just still figuring out where I want it to do the work—and where I want to.

100

Andy @amonimous.bsky.social · 21/07/2026

For years, we’ve relied on libraries and frameworks. They abstracted away common problems so we could focus on building the parts that made our applications unique. #buildinpublic #ai #softwareengineering #developerlife #coding #llm #futureofwork #productivity #devthoughts #programming

450

Andy @amonimous.bsky.social · 17/07/2026

Approach B is amazing. it feels like you’re moving at 10x speed. That’s also the trap. The more functionality you add without stopping to validate and refine, the more assumptions the AI accumulates. Soon you’re chasing logic through patches, conditionals and conflicting implementations

amazing.it

Amazing.it

Il portale https://www.amazing.it per gli amanti di Amazing.it.

000

Andy @amonimous.bsky.social · 17/07/2026

Approach A is slower in the moment, but much faster over the lifetime of the project. Each requirement becomes a solid foundation before the next one is added. That doesn’t mean there’s no place for moving fast. If I’m building a throwaway proof of concept, I’ll happily let the AI sprint ahead

000

Andy @amonimous.bsky.social · 17/07/2026

A working prototype communicates an idea far better than a document, a wireframe or a slide deck ever could. But when it’s your main codebase, speed without validation quickly becomes technical debt. In many ways, this isn’t a new lesson. The technology has changed. The engineering discipline hasn’t

000

Andy @amonimous.bsky.social · 17/07/2026

Approach A: Implement one requirement, review it, refine it, then move on to the next. Approach B: Keep feeding requirements to the AI, let it one shot it #AI #SoftwareEngineering #LLM #AIEngineering #Programming #CleanCode #SoftwareArchitecture #Developer #TechnicalDebt #GenerativeAI

440

Andy @amonimous.bsky.social · 15/07/2026

If you want to understand the value AI brings to your team, start with a baseline. Measure how you work today using metrics that matter—whether that’s DORA, SPACE, cycle time, lead time, quality or developer satisfaction. Then introduce AI at specific points in your SDLC and measure the impact.

000

Andy @amonimous.bsky.social · 15/07/2026

Did delivery get faster? Did quality improve? Did developers spend more time solving problems and less time on repetitive tasks? Did we deliver benefit for the Customer? That’s where the real value is. Don’t fall into the trap of producing more, in the hope that there is gold in it.

010

Andy @amonimous.bsky.social · 15/07/2026

One of the mistakes I see is introducing AI and then trying to work out whether it helped. This is the metaphorical equivalent of panning for gold in an endless churn of code. #AI #SoftwareEngineering #DeveloperProductivity #DORA #SPACEFramework #AIEngineering #LLM #Programming #Tech #SDLC

200

Andy @amonimous.bsky.social · 14/07/2026

Give an LLM too much detail, and you’ve already designed the solution. Give it too little, and it fills in the gaps with assumptions. Finding that balance is surprisingly difficult. I’ve found the most effective approach is to treat AI like a collaborative thinking partner

000

Andy @amonimous.bsky.social · 14/07/2026

When I have a new idea, I’ll spend time in an AI-assisted requirements session. I’ll ask questions, explore edge cases, challenge assumptions and gradually build a shared understanding of the problem. Set expectations from the start. I’ll explicitly say things like: * This is an early concept.

000

Andy @amonimous.bsky.social · 14/07/2026

* Keep the discussion high level. * Don’t dive into implementation yet. * Ask questions if something is unclear. * Help me explore the idea before proposing solutions. Without those guardrails, it’s remarkable how quickly the conversation jumps into architecture diagrams, database schemas and code

000