What AI tools do we actually use, and why?
Every week, there's a new think piece telling you which AI model is best. Most of them are written by people whose actual daily toolkit is a browser tab and a lot of opinions. So in this episode, we did something simpler: we told you what we, Giles and Theo, are genuinely using. In real work. For real clients. And — perhaps more usefully — what we've stopped using, and why.
Beside Ourselves is a podcast from Beside Partners — helping SMEs take bold, thorough steps in AI adoption. New episodes every two weeks.
First, the ChatGPT question
It's the one everyone starts with, so let's get it out of the way.
ChatGPT lit the fuse. We both signed up early, paid for subscriptions, and used it enough to have a view. That view is: it's fine, and fine turned out to be the problem.
Over time, it has become clear that OpenAI has aimed ChatGPT at the consumer market — and it shows in its outputs. There's an agreeableness to it, a tendency to validate rather than challenge, that technologists tend to clock fairly quickly. "Sycophantic" is the word that keeps coming up, and it's earned. Ask it whether your strategy is solid, and it will probably tell you it's great. Other models, in our experience, will tell you what you missed.
Giles is at the point of cancelling his subscription. The only thing keeping it alive is professional curiosity — useful to know what it's doing when clients ask. Theo hasn't opened it in months.
Where we've actually landed
Both of us have converged on Claude as our primary working model, with Gemini playing a significant supporting role depending on context.
Claude's quality of engagement is the differentiator. It doesn't just produce — it reasons, pushes back, and catches things the others let slide. Giles has built a set of Claude projects across different areas of his life and work, loading them with context over time. The more you give it, the more useful it becomes — a compounding effect that rewards consistency.
Theo uses Claude extensively as an editorial partner for his Substack. He does the writing himself, then brings Claude in to review, suggest, and challenge. Having tried the same setup in ChatGPT and Gemini, his verdict is clear: Claude, as a reviewer, is in a different league. It finds things the others miss.
Gemini earns its place for a different reason: ecosystem. When your clients are running on Google Workspace — and many of Theo's are — jumping to another model carries real friction. The integration matters more than the marginal difference in quality. For image generation specifically, Theo rates Imagen Pro as still the best available.
The model wars are fading
Here's the shift we're starting to notice: the question of which model is slowly becoming less important than the question of which ecosystem.
If your business runs on Microsoft 365, Copilot is already embedded in your tools. Google Workspace means Gemini is already there. The performance gaps between frontier models are narrowing. The switching costs between ecosystems are not.
That's not a reason to sleepwalk into your default — there are still meaningful differences for specific tasks, and knowing when to reach outside your stack has value. But for most businesses making practical decisions, the honest answer to "which AI should I use?" is increasingly: probably the one that's already woven into the tools you're paying for. The future conversation will be less about which model is smartest and more about cost, integration depth, and which applications are genuinely useful on top of the underlying technology.
The apps doing the real work
Two tools in particular are pulling above their weight in our daily workflows right now.
Granola (granola.ai) is our meeting notes tool of choice, and we're constantly asked about it. What sets it apart is that it transcribes from mic input — phone calls, in-person conversations, not just video calls. It captures everything, lets you query the transcript conversationally afterwards, and stays out of the way. Unglamorous, but genuinely useful.
WhisperFlow is Giles's newest enthusiasm, and he's fairly evangelical about it. Voice dictation that goes beyond basic transcription — it takes the natural, meandering way humans actually speak (false starts, self-corrections, thinking aloud) and renders it as clean, usable text. Emails, prompts, WhatsApp replies — all dictated. His description of it as "living the Star Trek life" is only a mild exaggeration.
Theo also gives a shout-out to Nous.ai, a knowledge base tool he's been deploying with clients. For teams drowning in SharePoint folders and scattered documents, it brings genuine order — and lets people actually find what they need.
A workflow worth stealing
On the subject of using AI without losing yourself to it: Giles's writing process is worth describing, because it's a genuinely useful model for anyone who worries about their voice getting flattened by AI assistance.
He starts with a fountain pen and paper. Thinks on the page, gets the structure down, then reads it back in digitally. Only then does Claude come in — not to write, but to challenge. Cross-reference against the back catalogue. Push deeper on the points that are undercooked. If Claude makes a fair point, Giles goes back to the pen, writes a new section, and feeds it back in.
The AI is an editor and a sparring partner. It's not the author. That distinction matters, and it's one we think more people should deliberately hold on to.
Has it made things better?
The only honest answer: yes, significantly.
Theo puts it plainly — operating across multiple clients, companies, and stacks simultaneously would not be possible at this level without these tools. Giles agrees. Things he thought were just slow, friction-heavy parts of the job — writing, coding, navigating complexity across multiple projects — have been meaningfully transformed.
We're aware of how that sounds coming from people who run an AI podcast. But we've both been around long enough to have watched technology hype fail to deliver. This one, used with some thought about how you use it rather than just that you use it, is delivering.
Our bottom line
Stop worrying about which model is theoretically smartest. Start paying attention to which tools you're actually reaching for, which ones earn their place in your day, and which ones you've quietly stopped opening. That gap tells you more than any benchmark.
We'd love to know what's in your toolkit — and what you think we're missing. Find us on LinkedIn or drop a comment wherever you're reading this.
