A couple of weeks ago, I was chatting with Claude about European flight rewards programmes, just a conversation on how I could improve one of my existing blog posts. Twenty minutes later, I had a functional Eurobonus calculator running on my blog. The interesting part? I never asked for an app. Claude just… built one.
This crystallised something I’ve been thinking about: we’re chasing the wrong metrics with AI tools. Every week, there’s a new benchmark, a new “GPT-killer,” another model claiming superiority. My take from this is that the most powerful model is useless if it doesn’t fit your workflow.
The most powerful model is useless if it doesn’t fit your workflow.
The AI Race
OpenAI announces GPT-5. Anthropic counters with Claude 3.5. Google throws Gemini into the ring. By the time you’ve finished reading this sentence, there’s probably a new model claiming to be 2% better at coding benchmarks.
For the average user, this creates paralysis. Which tool should you master? Where should you invest your learning time? More importantly, which subscription is worth it?
Here’s my take: Experiment with new tools, and revisit tools you have used in the past and consider how each tool fits into your workflow and workstyle.
With the vast array of AI tools available, we have a toolbox that assists us in our work. However, each tool has its own strengths, weaknesses, and learning curve. Furthermore, these tools are constantly evolving, which means the learning process is never truly complete.
This creates a dilemma: we need to find a balance between utilising what we already know and adapting to new tools.
I’ve found that the best approach is to dedicate about one or two hours each week to learning about new tools by reading blogs, watching YouTube videos, and simply testing them out. For the remainder of my time, I focus on using the tools I already know work well.
This strategy provides a good balance between leveraging the most modern tools available while avoiding the pitfalls of spending too much time switching and learning new ones.
In 2025, the tools below proved most valuable to me.
My AI Stack
Claude
I pay for Claude, and it handles all my major tasks. Reports, blog posts and (vibe) code that actually needs to work. Why Claude over the others? The tone. It writes like a thoughtful colleague, not a robot pretending to be human.
ChatGPT
I use ChatGPT for… everything else. Training plans, recipe searches, holiday planning, and quick calculations. Why split my usage? Simple economics and organisation. I don’t want to burn through my Claude tokens asking about pasta recipes.
Using two LLMs side by side is also a great way to compare results and assess how reliable and accurate the answers are.
Consensus
Consensus quietly revolutionised academic research. Real citations. Multiple format outputs (APA, Harvard, whatever you demand). It will provide you with a solid literature analysis in mere minutes.
Grammerly
I use Grammarly every day, and it has significantly improved my writing. It helps me write more clearly while speeding up my writing—an absolute no-brainer.
The only downside is that it can be a bit clunky at times, since it’s not native software but still has to work in every app I use. But overall, it works well enough.
NotebookLM
A great companion for learning is Notebook LM. Reading a book can often be boring. With Notebook LM, you can create engaging study materials in the form of videos and podcasts.
However, it’s important to remember that it’s not perfect, and you may encounter some incorrect information. I’ve actually enjoyed the process of discovering where the AI makes mistakes, as this can be a valuable way to learn as well.
Is Paying Worth It?
The Simple Economics
Start with the basic calculation: does it save enough time to justify the costs? If the paid programme saves you one or two hours of work a week, it’s probably worth upgrading to a paid tier. But there’s another layer most people don’t consider: privacy.
Does Paid Provide Better Privacy?
The uncomfortable truth: paying changes the contract, not necessarily the practice.
Here’s what actually happens with your data:
Free tiers: Your conversations can be used for model training by default. When you chat with ChatGPT or Claude for free, you’re essentially contributing to their training dataset. Your questions, your writing style, and your work patterns all potentially feed the next version of the model.
Paid tiers: By default, your conversations aren’t used for training with Claude Pro. ChatGPT Plus offers similar assurances. Anthropic (Claude) explicitly states that paid user data is automatically de-linked from your user ID before any use, and they employ privacy-preserving analysis tools to filter sensitive data.
However, there are important exceptions even on paid tiers:
- Conversations flagged for safety review can still be analysed
- Feedback you explicitly provide (thumbs up/down) may be used
- If you opt into specific programmes (like Trusted Tester programmes)
My Privacy Rules
Regardless of whether I’m paying or not, I follow these principles:
- Nothing genuinely confidential goes into any AI, paid or not. No client names, no proprietary methodologies, no sensitive financial data.
- Sanitise everything. If I need help with a client analysis, I generalise the scenario. Instead of “How should we assess cyber risk for Company X’s new cloud migration?”, I ask “How do you assess cyber risk for cloud migrations in financial services?”
- Use compartmentalisation. Serious work stays in Claude’s walled garden, where I pay for better privacy practices. Everything else—the recipes, the training plans—goes into the free ChatGPT.
- Understand what you’re trading. The productivity gains are worth the subscription fees, but don’t confuse payment with complete protection.
The paid tiers do offer better privacy practices, but your behaviour matters more than your billing status.
What’s your AI stack looking like these days?
Leave a Reply