February 8, 2026 · 11 min read

business AI software

2025 Vendor Audit: Every Tool We're Keeping, Killing, and Replacing with AI

Thank you Andrew and the Third South team for reviewing

There is too much noise right now about which SaaS companies AI will disrupt. We have inherited, tested, and cut dozens of vendors. We have messy credit card statements filled with legacy applications, developer tools, and infrastructure costs. I went through every single line item, from the sub-$5 utilities to the massive cloud bills. Who will survive 2026?

When you look at the survivors, a pattern emerges. We aren't replacing Expert Infrastructure (AWS, Postmark) or Expert Judgment (Lawyers, Accountants). They are safe.

We are killing middleware.

Middleware companies exist to make technical chores, setting up a server, scraping a page, monitoring uptime, slightly easier. They sell an "easy button." But AI has made those hard things trivial enough that the premium no longer makes sense. In 2026, we are paying for assets, not wrappers:

1. AI Replaceables

These tools are the walking dead to me. AI has lowered our barrier to entry so significantly that paying a subscription is now a tax on laziness. We can build internal versions more suited for us using pennies.

• Heroku: The biggest casualty. Heroku has been shaky, and PaaS fees are a bleed. In the past, managing a VPS was a headache. Now, we build our own solutions while leaning on AI to handle the configuration. Between the cost, the instability, and the likely shutdown, this decision was easy. We are migrating off. There is no reason for us to stay, and other PaaS have issues. Not to mention that with Heroku and some of these other companies that we're moving off of, the only thing we really need to pick up from them is the config. And with LLMs and AI, it's much easier to reproduce the config and have it be almost exactly the same, especially since all of these companies have good documentation. Their main offering is just configuration.
• Render & Vercel: Same story as Heroku. Great for initial prototyping a year ago, but as we scale, we are moving these workloads to our own bare-metal/VPS setup. It is not worth paying the markup when we can migrate an entire company in a single workday. For future prototyping, we will create a script with AI to get us set up.
• Framer: We used to pay Framer to build sites visually. We have replaced this with AI coding tools. Our developers can "visualize" and build websites so quickly now that the drag-and-drop abstraction is no longer a value-add. It is a limitation. Our frontend needs have evolved, and we are done with the guardrails of no-code.
• Yamm: We replaced this mail merge tool with AI agents. We still outsource the underlying mail servers, but the scripting and sending is cheaper, faster, and more customizable when we handle it ourselves.
• Better Stack, Binary Canary, & OnlineOrNot: We have specific preferences for on-call rotations. Instead of fighting a vendor's feature set, we will build a simple internal uptime monitor and ping us. AI makes writing these utility scripts trivial. We prefer a custom solution that integrates directly with our workflow. These are all good tools on their own- but for what we want, we need something different.
• Ngrok: Replaced with a custom solution using Cloudflare tunnels. It is easy to implement, and we are already deep in the Cloudflare ecosystem. Using AI for this felt like a breeze. Cloudflare's docs are OK, but they are well known enough to where it was easy to implement this.
• Hevo Data, Hex, & Mantle: We had a "modern data stack" that felt over-engineered. We are bringing data ingestion in-house. We used AI to pseudo-build the connections in preparation for these platforms; now we are just going to flesh them out and save the money. We learned a lot while building the connections here. So doing this will be extending what we already have documented. Mantle might be the most difficult to churn off of, simply because it connects to Shopify's API which uses GraphQL, and GraphQL is annoying to replace. More on this
• Metabase: They are pushing features we don't need. We prefer building lightweight, custom internal reporting tools. So we will. Especially when we bring data in-house. Visualization is already pretty easy to do. The main reason we liked Metabase is that it was easily connected to all of our other disparate data sources and made cross-visualization easier. We know the few things that we need to measure and getting fancy matters less to us now. We just don't need to think about it as much.

2. Death's Doorstep

These vendors are effectively dead to us. The service has degraded, they are over-engineered, or they simply annoy us. We are just waiting for the contract to expire or the migration script to finish.

• GitHub: We are actively looking to leave. The platform has become unstable. We might spin up a self-hosted solution, but our spend here is dropping to zero. GitHub Actions is the only thing keeping us, and in 2026, there is a 40%+ chance an action will misfire. If we can get past our distaste for Docker, we will build our own solution rather than migrating to a competitor. Really, the biggest thing for us is that GitHub is the best place for us to host our Git and use Git appropriately. Git has been around for a long time, and there are a lot of good things about it, but there are also a lot of terrible things about it. Looking at large files and blobs, GitHub is specifically really bad at now, and they've gotten worse. I've been seeing some Git competitors pop up over the past few years, and even another one that came out today. I think we are likely, if we find a really good Git alternative, it will just become a no-brainer.
• Acquire.com: We used to source deals here, but the quality has tanked. We are returning to relationship-based sourcing. We continue to use AI to curate our deal flow, and so far it is significantly better than wading through the noise on Acquire. The amount of noise from a couple of years ago is staggering. Sellers didn't want to listen to appropriate prices, dragged out the process, and had no one helping them figure out what made the most sense. The worst are tire kickers who don't actually want to sell their companies. These are tough to deal with. The noise doesn't make sense compared to what we could do: be more targeted with businesses, be more targeted with brokers, and continue having deal flow whenever people come to us to talk about selling—even if they're not selling to us. Helping people is always better. It doesn't make sense for us to be on Acquire. Building our own tools is much better.
• Notion: Ripped out entirely. We don't use it. The search feature was always poor. With AI, we can search our own raw docs much faster.
• Fathom & Plausible Analytics: Removed. We moved everything to Seline Analytics and are happy there. By the nature of our company, having all our analytics is one place and then feeding them into our custom-reporting is so much easier than having to pull in analytics from a dozen sources.
• Cursor: We experimented with this AI IDE. The composer model is really good, and we like it a lot. We still use Cursor. But in a year, will it still be the best IDE? We are far from married to Cursor...
• Firecrawl: We use this to scrape websites into JSON. It works great today, but scraping is core to our business. In 24 months, we will have built our own internal, AI-driven scraping infrastructure. This is a ticking clock. Only because it is core, though.

3. Cruft

Digital debris. Legacy accounts, forgotten subscriptions, and one-time purchases that clutter the P&L. We aren't replacing these with AI. We are straight cancelling.

• Linode (Akamai): Ripped out. No longer needed.
• Cloudinary: Legacy bloat. Cancelling and moving assets to AWS S3.
• IDrive: Legacy backup tool. Replacing with efficient, automated scripts.
• Intruder.io: I honestly don't remember why we pay for this. Cancelled.
• Gumroad: One time cost.
• DNSimple: Consolidating domains to Cloudflare/AWS. No longer needed.
• BaseDash: Used briefly for visualization, but unnecessary.
• Tuple: No one on the team uses it.
• Accelerator Keys: A one-time purchase we aren't renewing.
• Clearout: Used for a one-off email validation. We got the data; we're done.
• GoDaddy & Namecheap: Consolidating where possible, though some domains remain.

*Note: Colin said "I didn't even know that we had this cruft". Bloat bloat bloat, all of which will be deleted.

4. Unshakeable Cores

The bedrock. The AI revolution has zero impact on these because they sell commodities, not convenience. They are too cheap, too reliable, or too fundamental to replace.

• AWS: AI can write code, but it isn't a physical server farm. S3, EC2, and CloudFront are incredibly inexpensive for what you get. We migrated one of our businesses off of EC2 and to Heroku last year. This was the right move in the moment, but knowing what I know now, we would have kept the box we have.
• Cloudflare: They run our DNS and edge. They are excellent. We keep them.
• PlanetScale: We love them. We have several databases here and no intention of moving. Great team, as well.
• Postmark & Buttondown: Email is deceptively hard. Buttondown is core to our workflows, and Postmark handles transactional deliverability. We have no interest in rebuilding email infrastructure. Let them do it and we will ride their co-tails.
• Linear: It's good at what it does. We use it every day. Extensible enough to where we will think about replacing as we scale in five years, but we are happy with Linear today.
• Slack: We aren't building a chat app. It stays.
• Sentry: Essential for error tracking. No reason to remove this. I am not a fan of some of their AI-based features and moves to telemetry, but they are good at error tracking and we find them useful.
• 1Password: Essential for security. Could build something, but just not worth it.
• Google Workspace & Microsoft: We need email and docs. We aren't fighting this battle.
• QuickBooks: It's accounting. It works. We aren't disrupting tax compliance to save a few bucks.
• OpenPhone/Qua: Works well for our VoIP needs and 2FA.
• Zendesk: One of our businesses runs entirely on this. We can't rip it out.
• Help Scout: It's okay. We are building our own add-ons rather than upgrading, but keeping the core service. Similar to Linear in this regard.
• Snazzy Docs: A legacy deal makes this insanely cheap. Not worth the migration effort.
• Shopify, Squarespace, Wix: Necessary for specific business lines.
• Mercury: It's our bank. We are considering switching due to service issues, but obviously, we can't replace a bank with AI.

5. Irreplaceable Experts

High-judgment partners. We use AI to assist them, but we won't use AI to replace them.

• Legal Team: We are not replacing lawyers with AI. For high-stakes strategy and legal help, we need experts. However, for mechanical work (small contract tweaks), it doesn't make sense for us to pay the hourly rate anymore.
• Tax Team: Compliance is not a place for "move fast and break things." We keep our advisors and pay government fees as normal.
• Contractors: We still use humans where AI isn't ready or where a personal touch is required. That human element is vital across several of our businesses. Our current roster is great, and we're sticking with them. To us, all of these contractors fall in the expert category, and that is the only type of contractor we will continue to hire. For example, Karen has been in the art industry longer than I have been alive and has many irreplaceable connections. The same goes for Kristine. She is one of the best people we could have hired for social media. The same applies to Denisa. Although she is not an artist, I believe AI-generated articles will not match human-generated ones. She is a strong writer who puts in effort, knows what to write, conducts thorough research, and is excellent at expanding on topics we want to discuss. While she is not an expert in any of our businesses, she is an expert in writing, which is what we value.