We test a lot of AI tools. Part of our job is staying on the frontier — so when a client asks 'should we use X?', we have a real answer based on real usage, not just a demo. Here's what's actually in our recommended stack in 2026.
For AI models: Claude and GPT-4o
The model wars have settled into a practical duopoly for most business use cases. Claude (Anthropic) is our default recommendation for anything involving long documents, nuanced reasoning, or customer-facing text — its outputs are consistently more accurate and less prone to confabulation. GPT-4o remains excellent for multimodal tasks and cases where OpenAI's ecosystem integrations matter.
For automation workflows: Make (formerly Integromat)
Zapier gets the most press, but Make offers far more flexibility at a lower price point. For complex, multi-step workflows with error handling and conditional logic, Make wins handily. Zapier is still great for simple two-step integrations — but for anything we'd call 'real automation,' we reach for Make.
For AI agents: n8n + custom builds
For autonomous agent workflows, we typically build on n8n (self-hosted or cloud) combined with the Claude API for reasoning. n8n's flexibility and the ability to run it on your own infrastructure makes it the right foundation for agents that need to access internal systems. For simpler agent use cases, tools like Relevance AI have matured considerably and can be a faster path to deployment.
For document intelligence: Reducto and LlamaParse
Getting clean, structured data out of PDFs, contracts, and scanned documents used to be the hardest part of any AI project. Reducto and LlamaParse have both made remarkable progress here — they handle tables, complex layouts, and multi-column documents far better than the general-purpose OCR tools from two years ago.
What's overhyped right now
Most 'AI-powered' CRM and project management add-ons are thin wrappers around GPT that add cost without adding real capability. If the core product wasn't already excellent, the AI layer won't save it. Evaluate the underlying tool first; the AI features are rarely a reason to switch.
We also remain cautious about fully autonomous agent platforms that promise to 'run your business with AI.' The reliability and auditability requirements for truly autonomous business processes aren't there yet for most industries. Human-in-the-loop designs still outperform fully autonomous ones in most of our client work.
Ready to put this into practice?
We work with a small number of businesses at a time to make sure every engagement gets our full attention. Book a free discovery call to see if we're a fit.
Book a Discovery Call