Can someone actually explain to me what Sharepoint does?  If I worked in Microsoft’s sales department the best pitch I could give is:

“It’s the greatest, most versatile product that has ever existed. You can use it to run any complex system that your imagination could dream up.” This however would only be what I would pitch, not believe.

I’ve asked the question “ What does Sharepoint do?” to Microsoft sales staff, developers, and consultants. It always starts with something like: “Well… it’s, you know, like… a collaboration tool… BUT! It can do a ton of other stuff too”.

And that is the best answer I’ve gotten.

I’ve asked the same question of SAP vendors, Microsoft Dynamics consultants, and IBM Watson Cloud experts. The answer is always some amorphous, borderline ridiculous answer consisting of “well it does a lot of things” and “it greatly depends on the user”. This was not me asking rhetorical questions either. I was not trying to be glib, or overly clever, or even to pull some sort of #iamverysmart coup de grâce. I was trying to articulate what I do for a living by standing on the shoulders of “giants”.

You see, my company builds a “platform as a service” (roughly) type product as well. Something that could be more than one thing to more than one person. I struggle constantly with explaining that our product is better than anything else on the entire market. This is not a brag, nor a marketing ploy – but only because what we do is so niche that only 100 or so companies in the world might care. And that is not the game IBM, Microsoft, and SAP are playing. They are ultimately the owners of your software. Sure the configurations, the modifications, and the custom programming on top of these platforms is yours, but if they take the platform away, or stop supporting it, what do you really have left? It’s even tougher in “the cloud” business because then if your subscription runs out you’re dead.

I recently made a prediction to a friend who was starting a project with IBM. I warned them of the potential lock-in problem by making a prediction something along the lines of “They are going to tell you they can build it quicker and more effienctly with IBM Watson Cloud. No project ever runs perfectly, and when you finally step in to set things straight, you will find out you have zero leverage. They will simply say you are more than welcome to fire them, because they know you would have to build everything over from scratch”. My predictions were to no avail. No one ever gets fired for hiring IBM. And guess what happened? The only upside is that I get to say “I told you so” a little more often.

There is hope! There are other ways that platforms can be useful but also safe. One way is to use an open source platform, one that if at worst comes to worst, you can fire all your consultants and hire new ones, and the platform is still going to be around.

This is a little tougher with very niche enterprise products like ours, but we’ve done something a little different to combat my lock-in loathing: Our  products are OWNED by our clients. We sign a three year, non-exclusive agreement with our clients for support and maintenance,  and a traditional license fee is baked in. They get all the source code, and agree not to resell it. But if we don’t perform, or our clients want to go a different way, they get to keep the software and build on it themselves. We earn our right to be at the table by being the experts in a system we designed, working with their developers, adding new features, bringing our industry expertise to the conversation, and hundreds of other small bits of value. In this way we hope to be at the top of the renewals list in three years.

The idea of someone taking your software away from me is abhorrent. If your car company one day sent you an email saying that you now had to upgrade your fuel tank, and there was going to be a new subscription service if you wanted to keep using the same type of gasoline, you would riot in the streets. The model of software is not what is wrong here, what is wrong is the lock-in. Vendor lock-in is amoral. If there is no ability to keep something running, and there is no TRUE data portability option, then you are basically being extorted.

I get that as a business you are trying to maximize profit. I try to do the same thing. However I want to my product and my company to seen as sticky because we are valuable, and not because we would just be too painful to get rid of.

Previous ArticleNext Article
I help companies turn their technical ideas into reality.

CEO @Sourcetoad and @OnDeck

Founder of Thankscrate and Data and Sons

Author of Herding Cats and Coders

Fan of judo, squash, whiskey, aggressive inline, and temperamental British sports cars.

Image generation comparison from February 2026

I spend a lot of time generating images these days for presentations. My typical workflow is fairly scientific: I ask Midjourney to produce a relatively cute image of a frog, a toad, a robot, or some other vaguely anthropomorphic creature doing something related to the slide I’m about to present.

Once I get the image, I expand the background by about 90% so the character ends up in the corner of the slide. That gives me a nice, relatively clean area to drop text on top. Sometimes I use Photoshop to do the expansion. Sometimes Midjourney cooperates. ChatGPT is actually pretty good at this too. Nano Banana is… enthusiastic. It tends to try a little too hard right now.

That’s fun and all. But the more interesting comparison isn’t cute amphibians. It’s boring enterprise diagrams.

Recently I had to generate some architecture visuals for an RFP response. Rather than suffer alone, I decided to turn it into a model comparison experiment.

Below is a slightly simplified (but very real-feeling) prompt I used. The company is fictional. The buzzwords are not:

Create a clean, executive-level architecture diagram titled “Closed-Loop Member Intelligence Platform.”

The layout should be 16:9 and structured left to right with a circular optimization loop surrounding the system.

On the left side, show multiple member touchpoints feeding into the platform:
- Website (class browsing, account login)
- Mobile App (workout tracking, push notifications)
- In-Club Kiosks (check-in terminals)
- Wearable Device Integrations (fitness trackers)

Label this section: “Member Interactions Across Digital & Physical Channels.”

All touchpoints should flow into a large central hub labeled:

“Unified Member Profile & Real-Time Event Engine”

Inside the central hub, include:

- Web SDK
- Mobile SDK
- API Gateway
- Event Streaming Layer
- Clickstream Data Capture
- CRM Data Sync
- Identity Resolution Engine

Include a small sub-caption:
“Event-level data unifies anonymous visitors and active members into a single dynamic profile.”

From the central hub, arrows should flow to a right-side activation layer labeled:

“Real-Time Engagement & Orchestration”

Include these outputs:

- Personalized Workout Recommendations
- Dynamic Class Availability Messaging
- Triggered Retention Offers
- Membership Upgrade Campaigns
- A/B Testing & Experimentation Engine

Surround the entire diagram with a circular arrow labeled:

“Continuous Optimization & Revenue Growth”

Along the circular loop, include metrics:

- Engagement
- Conversion
- Retention
- Lifetime Value

Design style should be modern, minimal, and suitable for an enterprise SaaS presentation.
Use neutral tones with one accent color to indicate data flow.
Avoid clutter.
Make the architecture clear and readable for both technical and executive audiences.

Here are the results.

ChatGPT

Clear winner for “looks like a human consultant made this at 11:30 p.m. before a board meeting.” The text was incredibly legible. The layout was balanced. The hierarchy made sense. It genuinely looked like something you’d expect in a mid-market SaaS pitch deck.

I even did a reverse image search on some of the icons. No exact matches. That suggests they were generated rather than assembled from some common icon pack. Which is pretty cool.

Claude

Claude did something interesting. Instead of just giving me a static diagram, it generated a React application that rendered the architecture visually inside its canvas. I should have guessed this is what that nerd would do… in fact I did guess, but whatever.

That has upsides. I can tweak the code. I can modify the layout. I can version control it. That’s appealing to the nerd in me.

But technically it failed the homework assignent. It wasn’t what I asked for. I asked for a diagram image. What I got was a React app that displayed a diagram that I had to screenshot.

That said, I actually liked the aesthetic. It felt a little more “me.” Slightly less textbook. Slightly more product-thinking.

Gemini (Nano Banana)

The undisputed champion of 2026 in image generation, nano banana, was actually my least favorite of all of the designs. I think there’s something really weird about the arrows on the outside ring of this diagram. Why are there two arrows between “Engagement” and “Conversion”? Why are they different sizes? I did actually find a couple of exact matches when searching for some of these icons here, so so there might be some assembly on top of generation going here, but I cannot tell because these icons are so universal that it’s likely that that could just be a coincidence.

Midjourney

Ah, Midjourney. My current favorite for keynote frogs.

Completely and utterly useless for generating readable diagrams.

It’s phenomenal at stylized imagery. I’ve tuned it so much over time that it practically knows my aesthetic preferences better than I do. It’s like it’s been trained specifically to make amphibians that align with my personality.

The Omni feature (object permanence) is genuinely impressive. If you’re telling a visual story and need a character to look consistent across multiple scenes, or you’re creating a children’s book to convince your six-year-old that haircuts are not a violation of human rights, it’s fantastic.

But enterprise architecture diagrams? Nope, sucksville.

Wrapping Up

I was pretty sure that nano banana was going to run away with this one. Everyone I know works in banking or finance or medicine has been telling me how great the model is for generating diagrams and process flows. They’ve been raving about how things that were not possible three months ago are now completely durable with this model. It was a little bit of a surprise to see that my personal favorite was good old-fashioned ChatGPT. I think, for my personal use, I’m probably going to use Claude to generate diagrams because they’re a lot easier for me to tweak once they’ve been generated.

That said, I think this experiment showed that when I do this kind of work in the future, I’m just going to load up the same prompt in three different models and just pick the one I like the most. Some of it’s going to be personal tastes; some of it’s going to be how well the model interpreted the prompt, and some of it’s going to be the state of that particular LLM and its model on that given week.

And I’m going to stick to only using Midjourney for generating cute pictures of toads.