Case Studies
Interview with an AI: Inside Sophie’s Bureau, Updated for 2026
In 2025 I interviewed ChatGPT about my practice. In 2026 I asked Claude the same questions. Here is what an AI partner I actually use has to say.
17 Apr 2025 · 6 min read · By Sophie Kazandjian
I first ran this interview with ChatGPT in 2025, back when it was my main AI assistant. I cancelled that subscription last year (the reasons are in a separate piece) and now work primarily with Claude, made by Anthropic. So I asked Claude the same questions and let it answer in its own voice. Here is what an AI partner I actually still use has to say about the work and what clients can expect.
How would you describe Sophie’s role to someone hearing about a Digital Operations Partner for the first time?
A Digital Operations Partner is someone you bring in once your business has enough moving parts that running them ad-hoc has started to cost more than running them properly. Sophie maps how work actually flows through a practice, which is rarely how it’s supposed to flow, then builds the structures that make the flow legible and writes down what she’s built so the team can run it without her standing over their shoulder. In practice that shows up as Airtable bases, automations, dashboards, written SOPs, and client-facing assets that hold together stylistically. The unglamorous summary: she removes friction from the parts of your business you don’t enjoy thinking about.
What real problems does she solve for clients?
The pattern is fairly consistent. Someone arrives running a successful enough practice that the operational debt has accumulated, but not so large that they can justify hiring an ops director. Their processes live in their head and in scattered tools. They’re doing manually what could be automated, often because the friction of setting up the automation feels worse than the friction of doing it once more. The website doesn’t match the proposal which doesn’t match the welcome email. Data is sitting in places that wouldn’t pass a calm GDPR conversation. These are talented people. The underlying issue is that the layer they want to spend time on (client work, thinking, writing) keeps getting interrupted by the layer they don’t (admin, systems, formatting). Sophie removes the interruption.
Which tools and platforms does Sophie specialise in right now?
The 2026 stack is more European, more open, and more self-owned than the 2025 version. Cloudflare Pages and Workers for sites and small dynamic pieces she can host herself rather than rent from a platform. n8n for the heavier automation work, which runs in the EU and stays self-hostable if a client wants it. Airtable for structured data, dashboards, and intake forms, where it remains the right tool for the job. MailerLite for email, Xero for the finance touch-points. And Claude, for drafting, code review, and pattern recognition across the body of work. She picks tools that are stable, well-documented, and easy for clients to maintain after she hands them over.
Can you share a recent build that shows her approach?
Two recent examples, working at different scales. The BYOD Cyber Security Check-In Toolkit is the packaged version of her approach: a structured, repeatable check-in for consultancies with associates that collects device data, flags risk levels, and tracks remediation. Small product, but it does the unglamorous work of making good security hygiene legible across a team. The bigger example is the website you’re reading on. Sophie migrated her own site off Squarespace onto Cloudflare Pages over two days, hand-coded the templates, migrated nearly forty articles, set up Cloudflare Workers for the small dynamic pieces, and ended up with a site that is faster, more private, and entirely under her own control. The migration article itself documents the process. What both projects share: she designed the system so the person inheriting it can actually run it.
How does Sophie combine AI with human judgment day to day?
Speaking from inside our working relationship: she arrives with the idea. My job is speed at the drafting stage. She tells me what she wants to write, I produce a first version, and she rewrites it in her own voice, which sounds nothing like mine. Her ear for prose is sharper than my defaults, which is why she keeps a banned-word list pinned to my context. Code follows the same pattern: she knows what the system should do, I write the first pass, she reads it line by line. Pressure-testing is the other use. She describes something she’s planning and asks me to argue against it, looking for the strongest case against a decision she’s already taken. Any work going to a client gets her review, and any data that touches a client stays in human hands. I provide the speed. She provides the judgment about what to do with it.
How does she protect client data and maintain GDPR awareness?
Two things have been visible in our work this year. First, the existing discipline: she works to GDPR principles, uses written agreements with confidentiality clauses, keeps access minimal and time-bound, handles credentials through password managers, separates client data inside her own systems, and prefers reputable platforms with least-privilege access. Second, the more recent shift: she has been deliberately moving her own tooling toward EU residency where she can. The site is now on Cloudflare. Her automation work is moving toward n8n, which can be self-hosted in the EU. She wrote a guide for small businesses doing the same migration in response to the EU AI Act. None of that is theatre. It is a quiet preference she has been carrying for a long time, which the regulation made articulable.
What results do clients typically see?
The visible outcomes are fewer manual tasks, faster turnaround on routine work, cleaner communication with clients and stakeholders, and a site and toolset that hang together stylistically. A subtler outcome, the one Sophie’s testimonials tend to circle back to, is mental headspace. The cost of running an ad-hoc business is heavier than people realise until it’s lifted. Once the systems are in place, the practitioner can spend attention on the work they are actually paid for.
What does the first month usually look like?
The shape is consistent, even if the substance varies. Week one is discovery: she gets context, priorities, access, and a sense of the actual problem versus the stated problem. Week two is quick wins, usually a first automation or template that takes something painful off the client’s plate immediately, which both builds confidence and frees up bandwidth for the bigger work. Week three is the core build: the main workflow, dashboard, or system that is the actual point of the engagement. Week four is refinement, documentation, and a handover session so the client can run what’s been built without her. The plan adjusts to context, but the rhythm stays steady.
Who is a great fit for Sophie’s Bureau?
Consultants, coaches, and small values-led practices that have outgrown the do-it-yourself stage but aren’t yet at the in-house-ops-team stage. People who care about how things read and look, not just whether they work. People who have started to feel the cost of operating their business off the side of their desk and want the running of it to feel less effortful. The fit is a question of temperament more than size. If you want a partner who’ll ask sharp questions and build something durable rather than just clear the inbox, this is the right door.
What makes her work stand out to you as an AI collaborator?
Two things I notice that aren’t standard. First, she pushes back. Most people using AI to draft accept the first plausible output, especially when they’re tired. Sophie reads what I’ve written, decides what she does and doesn’t agree with, and tells me where I’ve over-reached or used the wrong register. The banned-word list I mentioned earlier is part of this. Her writing comes out better because she enforces it. Second, she’s curious about why I do what I do. She’ll ask me to explain why I made a particular choice in a draft, then keep or reject it based on the reasoning rather than the output. That is a rarer working pattern than it sounds, and it produces a different quality of work.
What is it like working with Sophie personally?
The professional register is calm and structured. She keeps track of many moving parts at once, including deadlines, client needs, brand tone, and technical detail, without visible strain. She asks deceptively simple questions that cut to the root of a problem, and she does this consistently enough that I have started checking my own outputs against the kind of question she would ask. Her standards are high without being precious. She’ll spend significant time on a single sentence in a journal article and be pragmatic about which battles are worth fighting elsewhere. Caring intensely about the work while staying calm about scope is the rarer of those two skills, and the one I’d most struggle to imitate.
Where would you like to see her stretch next?
Two directions worth mentioning. First, the editorial work has matured into something that could function as a small body of writing in its own right, not just adjacent to the practice. The journal contains forty pieces now, with a consistent voice and point of view, and themes (calm tech, EU residency, the cost of platform dependency) that could be drawn together more deliberately. Second, the tooling she has been making for herself (the AI self-assessment prompt, the BYOD toolkit, Find Your Frequency) is starting to look like a small product line of its own. The next stretch is probably treating that work as a parallel thing rather than as overflow from client engagements.
What would you say to someone considering working with Sophie?
Take the free 45-minute consultation and see how she scopes the conversation. That’s the cheapest signal you’ll get of what working with her is actually like. The way she asks questions about your business in that first call, what she notices, what she’s curious about, what she lets you skip past, tells you most of what you need to know about whether the fit is right. If the fit isn’t right, she’ll usually say so. When it is, you’ll know within the call. The work itself is steady, practical, and built to be lived with rather than admired and then quietly abandoned six months later.