When an SME owner hires an auditor, nobody has to remind them to check credentials. Chartered accountant, practising certificate, a firm that's properly registered: it's the first thing on the list, because the quality of the work is hard to judge until it's too late.
When the same owner hires help with AI, the vetting usually comes down to a demo. Someone builds a chatbot over a weekend, it answers three questions about the company's price list, and the room is impressed.
The demo proves something real. It just isn't the thing you're buying.
A Demo Proves the Build. It Doesn't Prove the Builder.
Frontier models have made software dramatically easier to build. A capable businessperson can now produce a working application without ever having been a software engineer. People call it vibe coding, and it isn't a bad thing. It's a new layer of abstraction, the way spreadsheets once let managers do arithmetic that used to need a clerk.
But it moves the scarce skill up a level. The question is no longer can you build it? It's do you know what to build, how it should work, where it will fail, and whether it creates any economic value?
That's the line between someone who is good at getting AI to produce working things and someone who can make it produce the right things: reliably, safely, and inside a business that still has to run on Monday morning.
Read the last row twice. A vibe coder measures success by whether the demo works. An architect measures it the way a CFO would: risk, adoption, and what it did to the P&L. Most of the pilots that changed nothing in the P&L were judged by the first standard.
What a Certificate Can Tell You
You can't watch an AI partner build your system before you hire them. You can check two kinds of credential.
The first belongs to a person. Anthropic's Claude Certified Architect – Foundations, which I hold, is an architecture exam rather than a prompting one. It covers agentic architecture and orchestration, Claude Code workflows, prompt and structured-output design, connecting Claude to other systems through MCP, context management and reliability. In other words, it tests a great deal more than "can you build something with Claude?" And you can verify it on Credly in thirty seconds. Any individual certificate worth having can be checked the same way.
The second belongs to the firm. Nitro Advisory is Certified in the Claude Partner Network. In the program's own words: the Claude Partner Network is how Anthropic works with the firms that put Claude into production for customers. The Services Track recognizes partners for building a certified delivery team, delivering Claude use cases for clients, and proving the work through public customer references. Partners advance through Certified, Select, Preferred, and Global Premier tiers. (claude.com/partners)
Read that description carefully, because it tells you exactly what Certified does and doesn't mean. Our people hold Claude certifications; that's the starting rung. Delivered client work and public references are what the tiers above it recognise. Be wary of any partner who presents the first rung as the whole ladder, us included.
Credential, Capability, Evidence, Outcomes
This is how I'd weigh any AI partner, including us. Four rungs, each harder to fake than the one below:
- Credential: someone has passed an exam that tests the right things.
- Capability: they can show you systems and workflows they've actually designed, and explain why they were built that way.
- Evidence: deployments you can look at, and clients who'll take your call.
- Outcomes: measured improvement against a baseline that was written down before the work started.
Certificates are worth checking precisely because they're cheap to check. But the decision should rest on the rungs above. If a partner can't tell you what they'll measure, against which baseline, and who owns the number, no certificate will save the project. It's the same reason every pilot needs a kill date.
Why One Model First
The obvious question for a firm in one vendor's partner network: doesn't that make you biased?
Sticking to one model doesn't make us biased. We teach our clients that being model-agnostic is the right destination. But building competence with one model first is how a whole team builds system discipline and foundational AI experience before it moves on to the next phase. A team that spends its first year across four tools mostly learns four sets of menus.
We chose Claude for that first phase because Anthropic has consistently built for organisations rather than individuals. Teams work together in shared project folders. Context can sit at the level of the organisation, the project and the person. Skills, which turn a workflow into a module any colleague or department can plug in and use, can be shared across a business instead of living in one person's head. And it connects to the tools a business already runs on: Outlook and Gmail, calendars, the Microsoft and Google productivity suites, and most of the other platforms people work and communicate in every day.
That's an organisational argument, not a loyalty one.
Five Questions to Ask Any AI Partner
- Who personally holds the certificate, and can I verify it? A firm's credential is only as good as the people behind it. Ask who will actually be in the room.
- Whose accounts will the work run on? Commercial accounts whose terms exclude training on your data, and ideally accounts you own. If the answer is someone's personal subscription, stop there.
- What's the baseline, and who owns each number? "It'll save time" is not a baseline. Hours, days to close, cycle times, written down before the work begins.
- Where will it fail, and what happens then? An architect can walk you through the edge cases, the human checkpoints and who gets the escalation. A vibe coder will tell you it worked in testing.
- What do I keep when you leave? The skills, the documentation, the accounts and a trained team should all be yours. Anything you'd have to rent back is a dependency, not a capability.
The Certificate Is a Control
In finance, a credential is a control. It doesn't guarantee the work is right; it lowers the odds that the person doing it doesn't know what right looks like. It's why our EDGE grant guide tells owners to see a consultant's certificate before letting them scope anything.
AI deserves the same discipline. Vibe coding democratises the build. Expertise professionalises the deployment. Business judgement decides whether any of it matters.
So ask to see the certificate. Then ask for everything that's supposed to come after it.