My2025 stack

Changes

Software has changed immensely over the course of this year alone, and honestly, so has my life. You know that feeling when your entire workflow feels like it’s held together with digital duct tape? That was me six months ago. So I did something radical: I simplified. Fewer tools, but each one earned its place by being genuinely versatile and reliable.

Search: Perplexity

Instead of bouncing between ChatGPT as my AI crutch and Claude (which I actually prefer for deeper work), I’ve landed on Perplexity. Here’s why: “Spaces”.

Think of Spaces as “projects” in Claude or “Custom GPTs” but with a crucial difference–you get to choose your model and the depth level for each task. Need quick facts? Lightweight model. Wrestling with complex analysis? Bring in the heavy artillery. The accuracy feels more consistent, and I’m not getting sucked into those rambling “chit chat” sessions that somehow eat an entire afternoon. Though, let’s be honest, it still happens sometimes.

My insight? When your search tool remembers context and lets you tune the intelligence level, research stops feeling like starting from scratch every single time. Not to forget, it can fact check my late night ramblings with sources I can double check.

Hot take? - ChatGPT is slowly becoming the ” Facebook” of language models. Change my mind…

Writing: Obsidian

I’ve evangelized iA Writer for years. Still think it’s excellent. But I outgrew it–needed more themes, better language support, and frankly, more horsepower. Enter Obsidian.

This thing is impossibly feature-packed. Community plugins everywhere, automatic knowledge graphs that show you connections you missed, integrations that just work. The graph view alone is like having a conversation with your own brain–suddenly you see the threads between projects you thought were unrelated.

Sure, it’s more technical. There’s a learning curve. But I’m a technical guy, and the payoff is worth the initial fumbling around. I only wish it would organize my documents for me, but maybe that’s coming sooner than I think.

If you are the type of person that also writes daily journals, this is for you.

Chat: T3.chat

The wildcard here is T3.chat. Created by Theo Browne (YouTube and Twitch fame), and it’s exactly what you’d expect from someone who’s as allergic to bloat as he is to bad developer experiences.

All the models as soon as they drop. Great UI. Great tools. Eight bucks for genuinely generous limits. Perfect for quick chats and simple image generation when you don’t need to architect the perfect prompt for twenty minutes.

Image / Video Generation: ComfyUI

ComfyUI’s new “native” integrations let you tap into models through their own API connections with transparent, fair pricing. This is obviously more technical–there’s a learning curve–but it’s gotten much smoother, especially when you consider the alternative is buying or spinning up massive servers just to experiment with decent models.

The concurrent rendering capabilities mean you can actually get serious work done when building complex image generation setups. And here’s the kicker: you can easily add language models into the mix, opening up all sorts of hybrid workflows.

That said, if you want to go deep and tweak every setting imaginable, you’ll want to look at services like RunComfy or ViewComfy for the full experience.

Warning: This rabbit hole goes deep. Budget your time, and wallet, accordingly.

IDE: Cursor

Cursor, despite some early communication hiccups about usage tracking, has found its groove. At $20/month, you get what feels like heavily subsidized access to powerful models, and the agent capabilities rival the Cline extension that initially held the code agent crown. However, you don’t pick the model. This was initially quite an issue, but not as much anymore.

Occasionally I’ll jump into Claude Code or Gemini in the terminal when testing new approaches, but unless you’re an expert at crafting agents and context (looking at you, Poul), your mileage will vary. That’s not me being dismissive–it’s just facts. Token generation doesn’t automatically make software that runs. It’s still engineers doing the work.

For twenty bucks, I get what I need. For now.

These days I’m playing with Godot Engine and Blender, trying to work out how to quickly create and test basic game ideas.

Privacy and experimentation: LM Studio

For privacy and experimentation, LM Studio is fantastic. Need a local chat interface that can occasionally serve as an LLM endpoint for coding? It runs everything locally and privately.

Perfect for those “I wonder what this model would say about my proprietary code” moments without sending anything to the cloud, or sensitive legal, or health matters.

Presentations: iA Presenter

When it comes to presentations, iA Presenter is still superior in every aspect that matters.

Making presentations without ever touching a button or dragging anything anywhere is genuinely liberating. You write, it layouts. You think, it flows. And here’s the magic: you can serve it as a responsive website for your audience and make changes instantly.

My takeaway? This is the fastest way to impress anyone with a presentation you did in 10 minutes. Just make sure the content is not total garbage.

Behind All This

The pattern here isn’t about chasing the latest shiny object. It’s about finding tools that amplify your thinking rather than fragment it. Fewer choices, more capability, less context switching.

Look, I’m not here to sell you on my exact stack. What matters is the principle underneath: your tools should make you feel more capable, not more scattered.

How many browser tabs do you have open right now? How many subscriptions auto-renewing for things you used twice? How often do you start a project in one app, get stuck, then lose momentum switching to three others just to finish a simple task?

This year taught me something brutal: the cognitive load of managing tools often outweighs their individual benefits. When you’re burning mental energy just remembering which app does what, you’re not creating–you’re administrating your own digital chaos.

The tools I kept passed one simple test: They either eliminated friction or added genuine capability. Everything else got the axe.

Your mileage will vary. Your needs are different. If chatGPT does all you need, awesome! Your tolerance for learning curves is different. But the question remains the same: Are your tools serving your work, or is your work serving your tools?

Don’t Let Tools Make You Lazy

Here’s the thing about Perplexity – it can be phenomenal learning partners if you don’t give in to laziness every single time. Create a Space, upload your files, URLs and context, set it to cite sources and contrast viewpoints, then force yourself to synthesize what you’re learning instead of just copy-pasting answers. The moment you stop thinking and start consuming, even the best AI becomes intellectual junk food. Use these tools to amplify your curiosity, not replace it.

When the tools disappear, and your ideas get done, that’s when the real work begins.