Co-host Episode Episode 02

Will Chief of Staffs
Survive in the Age of AI?

Two operators on the automations they've already built, the work they still trust themselves to do, and why the best Chiefs of Staff are about to become 10x more valuable — not redundant.

About This Episode

Is AI going to take the jobs of operators? It's the question on everyone's mind — from designers and engineers to finance and operations. Bea opens this episode with a confession: she automated herself out of 15 hours of work in a single month. The more interesting question isn't whether AI is capable of doing your job. It's what you do with the time you get back.

In Episode 2, Cecilia and Bea get into the specifics — the actual workflows they've automated, the tools they're using, the mental obstacles they had to overcome, and where they still trust human judgment over AI. More importantly, they make the case that the Chief of Staff role isn't dying. It's bifurcating. The operators who adapt will become dramatically more valuable. The ones who don't will become redundant. This episode is a practical, honest guide to which side you want to be on.

What We Cover in This Episode

The 15 Hours Bea Got Back — and What That Question Actually Decides

Bea opens with a provocation: she automated herself out of 15 hours of work in one month. Ceci immediately asks what she did with the time — and Bea's answer reframes the entire conversation. The question of what you do with recovered hours is precisely what separates operators who will thrive in the age of AI from those who won't. The episode builds from there.

How AI Has Already Changed Day-to-Day Operator Work

Both hosts take stock of how their workflows have evolved over just two years. Ceci traces the shift at her previous company from manually copying data across spreadsheets and writing summary updates herself, to using the Claude extension for Google Sheets to automate the entire weekly reporting process. The key reframe: stop measuring your work in hours logged and start measuring it in output and impact. Tasks like copying data between tabs aren't value-add — thinking about what the metrics mean for the business is.

Three Automation Use Cases That Changed Everything

Bea walks through the three areas where Claude Code has made the biggest difference in her role. First, the investor dashboard: a previously four-day process involving an external financial analyst, a third-party platform, and multiple rounds of back-and-forth is now a self-updating HTML web app that pulls automatically from a Google Sheet. Second, legal agreements: what used to require expensive outsourced counsel is now handled through Claude for the majority of contracts, reserving external counsel only for strategic agreements. Third, customer support: roughly 50% of Praktika's support function — predominantly refunds and cancellations — is now handled by an AI agent built with an external automation company and integrated into Intercom. Bea is direct about the implication: entire software categories have been made redundant overnight.

Building Web Apps Instead of Buying Software

Ceci argues that custom web apps represent one of the biggest structural shifts coming to how startups operate. She gives the example of sales commission tooling — a choice that used to be either a fragile, unmaintainable spreadsheet or a £15–20k enterprise software purchase. With Claude Code, you can now build a fit-for-purpose tool that understands your specific SKUs, your SPIFF structure, and your AE profile. It's not a 30-minute job, but investing two or three days delivers a tool that saves real money and fits your exact context. The shift in mindset required: stop reaching for off-the-shelf software and start asking what can be built.

The Mental Obstacle: Learning to Trust the Machine

Both hosts are candid about the learning curve — and both locate it in the same place: mental, not technical. Ceci's breakthrough came from building things with no stakes attached, starting with a personal website and an app for friends. Getting comfortable with HTML output and understanding how to interact with code, host things, and configure deployments took practice in a low-consequence environment. Bea's practical framework for working with Claude on web apps: one change at a time, precise language about assets and positions, and a recognition that some Excel operations — particularly structural row and column changes — are still better done by hand before asking Claude to layer formulas on top.

What We Still Trust Ourselves More Than AI

The hosts are equally honest about where they haven't delegated — and why. Bea keeps her hands on weekly cash flow management personally. When something is existential to the company, the responsibility doesn't transfer to the tool, and the judgment calls embedded in reviewing bank movements — like spotting software overspend and deciding whether to upgrade to an enterprise plan — require a human opinion, not just data processing. Ceci holds onto board executive summaries and sophisticated revenue narratives for similar reasons: the nuance required to convey a message that the data alone doesn't tell demands a level of contextual judgment that AI doesn't yet match. Both acknowledge this will evolve — but haven't delegated it yet.

Will the Chief of Staff Role Survive? The Three Irreplaceable Parts

After five mini panic attacks (Bea's count) and some serious reflection, both hosts arrive at the same conclusion: the role survives, but it has to be understood correctly. Ceci identifies three components that can't be automated. The first is context distribution — a significant part of the CoS role isn't task completion, it's making sure the right information reaches the right people at the right time to keep the organisation moving at speed, which requires constant human judgment about who needs to know what. The second is execution accountability — someone has to be the person who genuinely doesn't drop the ball and who understands in real time when priorities have shifted. The third, and most fundamental, is problem scouting: a good Chief of Staff doesn't wait for problems to be assigned to them. They move through the organisation, surface issues nobody has named yet, and figure out how to solve them. AI can help execute the solution. It cannot find the problem.

Who Should Own AI in a Company?

Ceci argues that while every function leader should be a champion of AI within their domain, the Chief of Staff's relationship with AI is structurally different. Where a marketing leader asks "how can AI make our campaigns better?", the CoS should be asking "how can AI increase the surface area I can cover across the whole organisation?" The maintenance burden of processes built and delegated drops dramatically, which means a CoS with strong AI fluency can manage 10x more surface than before. Bea's view is simpler and more emphatic: everyone should get their hands dirty. Not doing so is laziness. The tools are accessible, the barrier is low, and the downside of not starting is becoming obsolete.

How to Get Started (Even If You've Never Written Code)

The hosts offer practical starting advice for operators who haven't yet made the leap. Bea's framework: audit your tasks for a week, map them by impact level, start automating the lower-impact ones to build fluency, then work up to the manual work that consumes the most time. The interaction model with AI doesn't require hard technical skills — it rewards clear, thorough communication. Ceci's addition: use Claude to explain things to you step by step, as if you're five years old, before asking it to build anything. Maintaining a sense of understanding and agency over what the tool is doing — rather than just executing its output blindly — makes the whole process less daunting and keeps you in control.

Frequently Asked Questions

Will AI replace Chief of Staff roles?

The role won't be replaced — but it will bifurcate. Chiefs of Staff who embrace AI and use it to expand their operational surface area will become significantly more valuable; those who remain attached to manual, task-based work will struggle to justify their role. The irreplaceable parts of the job — distributing context, maintaining execution accountability, and surfacing problems nobody else has named yet — are inherently human and cannot currently be automated.

What are the best use cases for AI automation in an operator role?

The highest-impact automations tend to be recurring, data-heavy, and currently manual: investor updates and reporting dashboards, legal agreement review for standard contracts, customer support routing and resolution for common request types, weekly metrics tracking and summary generation, and financial modelling. The pattern is consistent: if a task involves pulling structured data from known sources and producing a predictable output, it can almost certainly be automated.

How do you get started with AI automation if you're not technical?

Start by auditing what you actually do. Spend a week logging your tasks and categorising them by impact and frequency. Begin automating lower-stakes, repetitive tasks to build comfort and fluency. Use personal projects — a website, an app for friends, anything with no professional consequences — to experiment. Ask AI to explain things step by step rather than just doing them for you. The communication skill required to interact effectively with AI is within reach of anyone; the main barrier is mental, not technical.

Should Chief of Staffs be the AI champions in their companies?

Yes — but with a specific angle. Every function leader should champion AI within their domain. The Chief of Staff's role is to champion AI at the organisational level: identifying cross-functional automation opportunities, reducing the maintenance burden on processes across the company, and modelling what 10x operating leverage looks like in practice. Their broad context and lack of a single functional home makes them uniquely positioned to drive AI adoption holistically rather than within a single lane.

Episode Transcript

Introduction

Bea [00:09]

Welcome to Top of the Ops, the podcast where we have real conversations about what happens behind the scenes of startups. I'm Bea, former VC at Lakestar, now Chief of Staff at Praktika.

Ceci [00:19]

And I'm Cecilia, former VC at Talis Capital and now FD at PortalOne. We both made the jump from VC to operators a couple of years ago, and today we're talking about something that's on everyone's mind: is AI going to take our jobs?

Bea [00:34]

I automated myself out of probably 15 hours of work this month.

Ceci [00:41]

Wow. What did you do with the 15 hours?

Bea [00:43]

That is precisely the question that decides which operators will survive and which won't. And that's what we're going to explore in this episode.

How AI Has Already Changed Our Work

Ceci [00:53]

Talking with a lot of people working in startups these days — designers, software engineers, people in finance — everyone is asking whether AI is making their job redundant. If we look at what Chiefs of Staff actually do on a day-to-day basis, yes, there are so many small tasks that can be automated. I was thinking about my previous role at Seqera. One of my objectives when I joined was to make us a far more data-driven organisation than we'd previously been. At the start, you're gathering data, building the whole infrastructure and structure from scratch. There's loads of manual work, and that kept me busy for days. But when you step back and stop thinking about your work in terms of hours put in — and start thinking about the output — that's not really value-add. The special sauce of a Chief of Staff or operator is around ideating a workflow or a process, not copying data from one Google Sheet to another. You want to be thinking about what the core metrics are, how to slice them, how to present them, and what they actually mean for the business. If you can shift your time to thinking about the impact those metrics have on the business and how to drive changes, you absolutely should.

Bea [02:16]

I'd like to take a step back and ask you, Ceci — tell me about a specific task. When you started in your Chief of Staff role, how were you doing it compared to how you were doing it a year later?

Ceci [02:32]

It was actually two years of change. When I first joined, I was just asking ChatGPT questions. What changed for me most was spreadsheets — a hundred percent. I don't write a formula anymore. I know how my spreadsheets work, but it's just so much faster to use Claude to build them. And then, about a month ago, with Cowork, you can actually teach AI to replicate your workflows. If you're moving between tabs, copying one piece of data to another and building a report that way, you can turn it into a reusable command that runs on a regular basis. That shaves off hours of work that aren't really value-add. What about you? What's been the big game changer?

Bea's Three Major Automation Use Cases

Bea [03:21]

There are probably three main areas where I'm using Claude Code now. The first is investor updates. We used to have an investor dashboard on a piece of software called Aleph, tied to SharePoint in our accountant's database. I'd have to ask the financial analyst at our accountants to update the SharePoint, which would then flow through to Aleph, which then had to be tuned manually — usually with a couple of rounds of back-and-forth. That whole process, from start to finish, would take about four days. Then I'd summarise the product updates by listening back to previous town halls, draft the body of the email, attach the dashboard, and send. Now I've replicated the whole thing as an HTML web app. It pulls from a Google Sheet, automatically picks up last month's data, and updates itself. You can layer Cursor on top to pull data from different sources and update the Sheet automatically too, so everything runs on a schedule. Of course, you have to catch certain errors — connections break sometimes — but it's very manageable on a monthly basis, and it saves a huge amount of time on data processing. Side note: you can see why software stocks took a hit. A company that was providing a financial dashboard product is now completely redundant. It's genuinely better to build your own Claude Code automation.

The second use case is legal agreements. We used to spend a lot on outsourced legal counsel. Now I use Cowork, drop in the legal agreement, and only escalate to our counsel for strategically significant deals. For the majority of agreements, Claude is perfectly sufficient. The third is customer support — we've automated roughly 50% of our customer support function through an AI agent. We didn't build it internally; we worked with an external automation company, since customer support isn't our core business. But they've been great at implementing the automations based on our requirements.

Bea [05:55]

Oh — my cat Luna is going to join us. Anyway. The agent handles the bulk of our refunds and cancellations, which is probably 65% of what comes through on the support side. It's gone through several iterations and is now in a good state. It automatically routes tickets to the right inbox based on the topic, covering around five different categories of refunds and a few categories of cancellations. General questions are answered from a knowledge base that someone keeps updated — shout out to Daniel. And tickets that can't be resolved by AI get routed to a separate inbox for human agents. There's also an escalation path: if a customer at any point says they want a human, they get one. It's a bit of a mix, and yes, there's some customer frustration at the start — but our customers are genuinely lovely, and once their query is resolved by a human, they flip back to being satisfied pretty quickly. Alongside all of that, financial modelling and Excel — absolutely, as you mentioned.

Building Web Apps Instead of Buying Software

Ceci [06:02]

Web apps, as you mentioned, are what's going to change a lot in how startups operate. At my previous company, we went back and forth on whether our very complex Excel spreadsheet was a maintainable tool for calculating sales commissions. The alternative was paying £15–20k for an external commission tool. I was playing around with Claude Code and realised: you can build a great interface that's exactly fit for your use case. It understands your specific SPIFFs, your product SKUs, the types of AEs you have. It's probably not a 30-minute job — if you invest two or three days building it properly, the savings are significant and you end up with a tool that's perfectly calibrated for your company. The shift is about creativity: rather than identifying a need and immediately looking for a software provider to configure, ask what can be built. What can be built often fits better and costs less.

The Learning Curve: Mental, Not Technical

Bea [07:15]

One more question: was there ever a moment where you hit an obstacle implementing automation — where you realised your prompting wasn't working and you needed to learn how to interact with the tools differently?

Ceci [07:36]

At the beginning, the biggest jump for me was moving from using AI to ask questions — where it tells me how to do something and I do it myself — to understanding that I can actually interact with code, build a web app, host it, configure it. It was more of a mental obstacle than a technical one, because I'm not a coder. The tool would give me a bunch of HTML and I didn't know what to do with it. What really helped was playing around for my own personal purposes with no consequences — I made a personal website, built things just for myself. That was a great learning environment. Once I got comfortable there, I started building things for my company.

Bea [08:31]

For me, when I started with web apps, it felt more intuitive than working with Excel. A few things I learned: it's much better to give Claude one change at a time rather than listing five at once — that can cause confusion. Second, there's a specific language when working with images or assets in Cowork. If you want a logo replaced in a specific position, you have to say exactly that, or you'll get something completely different. And third, when working with Excel in Claude Code, I find that certain structural changes — moving rows or columns — use a surprising amount of computing power. For those, I've found it's often better to make the structural change myself and then ask Claude Code to build the formula on top. I don't fully understand why, but that approach works better.

Ceci [09:50]

A question back to you: you've spent a lot of time in Excel. Is there anything you're still better at doing yourself, especially in Excel? Or where's the aha moment been for Claude specifically?

Bea [10:05]

I recently switched to a MacBook — the first time I've abandoned my Lenovo ThinkPad after about 13 years. On Mac, Claude is now genuinely better than me at Excel. But I'd argue I'm probably still faster on Windows Excel, because I'm faster than waiting for the tokens to process in Claude Code. That's not a flex — if you've spent a few years in investment banking, you're probably in the same position. But it is enormously helpful now that I can't, for the life of me, do a keyboard shortcut on a Mac. Rather than getting frustrated and doing everything with my mouse, Claude Code is a great solution.

Ceci [10:57]

Welcome to the Mac world. What Claude gives me on financial modelling specifically is a version that's maybe 70–80% done, a great environment for thinking and iterating, and the ability to add my own layer on top. Because I tend to think by doing, having a model produced quickly means I can iterate so much faster. I don't have to build it first before realising I want it structured differently. That iteration speed is transformative. I built a full financial model for a consumer company in about three hours. Me and my Mac could not have done that before.

The Seqera Data Story

Bea [11:48]

Tell me more about the data infrastructure work you did at Seqera. I think it would be genuinely interesting — even for me.

Ceci [12:05]

I had to build a bookings tracking spreadsheet because we had live ARR tracked in our CRM, but nowhere to track bookings or committed ARR. At the start, it was entirely manual — I was opening deals, reading the incremental ARR, and pasting values in. ChatGPT helped me set up formulas and structure, but the data entry was all me. From that, I was building summary tables and writing update messages. A big part of that spreadsheet was a weekly tracker: every week, you capture changes across all the core metrics, and it gets presented to the leadership team. Every week, I'd copy the previous week's metrics into a new row, then hunt across several other spreadsheets to find the specific numbers I needed — all hard-coded because it was a week-on-week comparison. That took real time. Once I found the Claude extension for Google Sheets, I walked it through my exact workflow — what cell in which spreadsheet and which tab to pull from — and taught it to do exactly what I was doing. By the following Monday, the sheet was already populated and ready to present. And then you can take that, pass it to Claude, and ask it to generate a few bullet points summarising the changes. Before, I was writing that myself. Now I review it, add context for anything I specifically want to flag, and move on. It's a bulk of work I no longer need to do.

The Customer Support Challenge

Ceci [14:05]

Tell me more about customer support. This is one of the areas every leader in every company has been asking about — surely there's a lot we can automate. But it's a double-edged sword: you don't want to upset your customers with a dumb agent that doesn't understand what they're asking. How did you navigate that?

Bea [14:24]

I'm not sure we've fully cracked it yet — I'm not sure anyone has. What helped was getting in touch with a company that had spun out of a Lakestar portfolio company. They'd built customer support automation for that company, then spun out to serve other clients. We didn't qualify for the larger enterprise-focused players like Decagon — our volume was still relatively small — so this was the right fit for where we are. They handled the implementation based on our decisions, which was the right call because I didn't want to use our developers' time on customer support when they should be focused on the product.

We use Intercom plus an AI agent that functions as a seat within Intercom. The agent handles the bulk of refunds and cancellations — roughly 65% of incoming requests. Over several iterations, the routing is now clean: tickets go automatically into the right inbox based on topic, covering around five subcategories of refunds and a few of cancellations. General questions are answered from a knowledge base that's kept up to date. Tickets that can't be resolved by AI get routed to human agents. And at any point in the conversation, if a customer asks for a human, they get one immediately. There's definitely some frustration at the beginning of the AI-handled journey, but our customers are genuinely kind, and once their issue is actually resolved, they come back around quickly.

Will the Chief of Staff Role Survive?

Ceci [16:51]

We've just spent a good chunk of time talking about how amazing AI is and how many of our processes it's replaced. Going back to the opening question: what do you think actually happens to Chief of Staff roles and operator roles going forward?

Bea [17:25]

This has caused me about five mini panic attacks over the last few weeks. I was sitting there thinking: wait, am I needed here? If I've automated so much of what I do, surely I'm about to be fired. Classic imposter syndrome. But I actually think Chiefs of Staff exist for two fundamental reasons. First, because their principals can't be everywhere — and depending on the type of CoS, they're either making decisions on behalf of the CEO or managing the operational layer of the company. Second, organisations of every size experience coordination problems that require someone with broad context to direct and route. That's much more about emotional intelligence than automation. Right now, Claude doesn't have access to all the systems and context within a company — and even if it did, many of the conclusions you draw aren't in the data. They're in the unconscious processes, the unspoken tensions, the things that haven't been written down yet. So: big spoiler, I think we still need humans. But what do you think?

Ceci [18:57]

I agree. Someone at my previous company called me "the glue." I didn't understand it at first, but when you think about what a real Chief of Staff does, there are three parts that are genuinely irreplaceable by AI.

The first is context distribution. A lot of what you're doing isn't a task — it's making sure the right information reaches the right people at the right time so the organisation moves at speed. It's figuring out who needs to know what and when. That requires constant judgment calls.

The second is execution accountability — being the one person who genuinely doesn't drop the ball. Everyone drops the ball all the time because we're all juggling too many things. You need the one slightly annoying person making sure things actually get done, on time. And that can't be fully automated because priorities shift within hours, and you need someone who has that real-time read on where things stand.

The third is the most fundamental: a good Chief of Staff doesn't just execute what they're told. They move through the organisation, scout for problems that haven't emerged yet, spot processes that can be improved — and then figure out how to fix them. You can absolutely use AI to execute on solutions. But you can't use AI to discover the problems that nobody has identified yet.

Bea [20:52]

Especially if nobody even knows the problems exist. That's really important. Do you think the Chief of Staff should be the internal AI champion? And if so, why?

Who Should Own AI in a Company?

Ceci [21:10]

I think there should be several champions. Any engineering leader who isn't championing AI right now is missing a lot. Every function leader should be asking what AI can do for marketing, product, engineering, everything. But the Chief of Staff's angle is different. They're not asking how to make one specific function better — they're looking at the whole company and thinking about how to use AI to increase the surface area they can cover. The CoS model is: get parachuted into a problem, figure it out, create something, then either maintain it, delegate it, or move on. AI dramatically reduces the maintenance burden on everything they build, which means a CoS with strong AI fluency can cover 10x more surface than before. They should be champions of AI adoption for the entire organisation — not just one function.

Bea [22:34]

I think everybody should get their hands dirty and start playing. If you haven't started yet, you will soon be obsolete. There is an AI use case for every function, and it's genuinely easy to get started — you just log on and try things. If something doesn't work, you refine the prompt and try again. If somebody isn't trying, honestly, it's just laziness. And you should be excited that there's something taking the menial tasks off your plate — because so many times I get frustrated that I can't get into the strategic layer because I'm so buried in the operational detail. This is the opportunity to have a thinking Thursday. To look at the data, look at the problems, and actually figure things out.

What We've Struggled to Crack

Ceci [23:42]

Is there anything you've struggled to crack with AI in your day-to-day? Something that hasn't quite worked the way you work?

Bea [23:54]

Yes. We maintain a weekly sheet with all the business's cash inflows and outflows. Praktika is one of the most rigorous companies I've encountered when it comes to cash management — it's optimised to the cent, and I genuinely love it, because we will never run out of cash. But for that I just trust myself more. It's such an existential risk if we get it wrong, and I'm not ready to hand it over yet. Some of the data collection — subscription management, bookings, UA spend — I'm happy to automate, and that's actually on my list for next week. But the bank account movements, I want to review myself. Because I'm going to form an opinion about whether we're overspending on a given tool and whether, say, we should move to an enterprise plan for Claude Code. The responsibility doesn't transfer with the task. How about you?

Ceci [25:26]

Board executive summaries and sophisticated revenue narratives. When you're writing an executive summary for the board, it's so nuanced — you want exactly the right word, you want to convey a message that's often not coming directly from the data, because you have context that the data doesn't capture. That's where I still trust my own storytelling over AI's. Any project that's highly contextual, highly nuanced, and genuinely high-stakes — that's where I have the most resistance to fully delegating.

Bea [26:05]

And if I can add one more point — actually in contrast to that: one of my most humbling moments was when I did the investor update email body. I pulled all the town hall transcripts, dropped them into Claude, and asked it to write a product update summarising what we'd built in the last month. Claude produced this great summary. I pasted it in, sent the update, and a couple of days later got back responses saying it was the best qualitative investor update we'd ever sent. Everyone loved it. And it was the first time I'd just let Claude do it entirely by itself. I thought I was a good writer. Turns out Claude might be better. That was a slightly depressing moment.

Who Thrives and Who Struggles

Ceci [27:03]

Thinking about the future of the role — who is the type of Chief of Staff that actually thrives in the age of AI, and who is the one that struggles?

Bea [27:15]

If you can't automate 50 to 80% of the work you currently do, and you're still over-indexing on manual pulls and tasks that could be automated, I think you're in trouble. But if you flip your mindset toward automation and then abstract upward into strategic work, you become 10 to 20 times more efficient and more valuable. So if anything, the Chief of Staff who doesn't adapt becomes obsolete, and the Chief of Staff who becomes the company's AI champion becomes far more valuable. It's a real opportunity — but it is a risk if you stay attached to the menial.

How to Get Started

Ceci [28:05]

For those who haven't played as much with AI yet — the models have evolved so much that walking into Claude Code for the first time without knowing how to code can feel daunting. What would you recommend as a way to just get started?

Bea [28:25]

You can't just open Claude Code and expect something great to happen immediately. You have to be self-aware first — audit everything you do for a week. Write it down, mark it in your calendar, just capture it. Then understand which tasks are highest priority and highest impact, and which are lower-impact but time-consuming. Start automating the lower-impact ones to build fluency, then work up to the bigger ones. But the starting point has to be an honest audit of where your time actually goes. Once you've done that: try something, low stakes first, like building a personal website. Get your foot in the door, then upgrade to more complicated tasks. The communication skill required to interact with AI isn't a hard technical skill — as long as you're thorough and precise in your communication, you'll get what you need.

Ceci [29:28]

For me, one thing that really helped was asking Claude to explain things step by step — as if I'm five years old, just walk me through it. I also use Projects heavily in Claude and give a lot of context upfront, which makes it much easier to get what I want. When I'm doing something technically challenging, I always ask it to explain what it's doing and why, rather than just executing. That keeps you in the driver's seat — you still understand what's happening and why, rather than just running whatever the AI produced. That sense of agency really calmed me down when I was doing things like setting up a database for the first time.

Bea [30:18]

And it is such time well invested. This isn't going away. There was a LinkedIn post recently showing that only about 3.2 million people in the world are actively using AI for automation — against the entire working population. We're in an extraordinarily privileged position to be seeing this happen in real time and applying it directly to our companies. We are, in a small way, determining how a lot of the world will work one or two years from now. That is a great investment of time. And one last thing for any Chief of Staff listening: double down on what makes you irreplaceable. Be in the rooms where decisions are made. Catch the unconscious biases. Understand what the company's real priorities are versus the vanity metrics it might be chasing. Connect the dots across functions. That is high-fulfilment, high-impact work that you should protect and expand — not deprioritise in favour of tasks that a machine can do better.

On Time, Thinking, and Being Human

Ceci [31:41]

When your time is sunk into doing things over and over, and you feel overworked, that's when you have the least cognitive load available to think about the hard questions. And the real value of a Chief of Staff is also being the CEO's sounding board. You can't do that when you're buried. Your time is so much better spent when you take a step back and spend one or two hours thinking about a genuinely consequential decision — are we doubling down on direction one or direction two? You need the same kind of uninterrupted thinking time we used to have in VC. You can't provide that value to a company if you're overloaded with things that should have been automated months ago.

Bea [32:29]

And what a privileged position to be in — to just use your brain.

Ceci [32:34]

It's what makes us human.

Rapid Fire

Bea [32:35]

Let's close with some rapid fire questions. I'll let you go first.

Ceci [32:41]

Okay. I hope this doesn't happen — but if AI was banned tomorrow, what would you miss most?

Bea [32:48]

I'd miss Claude. That's it. Only Claude. I could do without ChatGPT. I could do without Cursor. I could do without Lovable. But not Claude. Claudio is our best friend. Your turn — what are you still better at than AI?

Ceci [33:05]

Controversial, but writing. I write fast, it doesn't take much effort, and I'm annoyingly precise about my style and the subtle sarcasm in it. Claude hasn't learned that yet. So: writing, for now. Next question — if Anthropic or OpenAI could hear us right now, what would you ask them?

Bea [33:27]

I'd ask about data safety and security. I have genuine concerns about the volume of information these LLMs are ingesting and what protections are actually in place. There have been a few controversial incidents where confidential information was surfaced. Before I dump our entire financial history and customer data into a model, I think companies have a duty to make sure that data is protected — and right now it's still quite unclear on the providers' side. What about you?

Ceci [34:07]

I would ask them to buy Salesforce and make it possible to interact with it through AI models rather than through Salesforce itself. That would be fantastic. I know Salesforce is a huge part of how B2B SaaS companies operate. We used to use it. It's very 2015.

Bea [34:18]

I thought you were going to say you hope they buy Salesforce and shut it down.

Ceci [34:23]

That too — but step one is enabling people to interact with it through AI. Shutting it down can be step two. And absolutely no offence to Salesforce or LinkedIn — we just think there are some things these platforms could do better. Let's be honest.

Bea [34:55]

In conclusion: Chiefs of Staff will survive — but only the good ones. And "good" is about to be redefined as people who can operate at 3x, 10x, 20x speed with AI. Let AI handle the grunt work while the Chief of Staff provides judgment, context, and the things that can't be written down. This week: take one recurring task from your workload and automate it. Block a few hours to experiment with a frontier AI tool. And have a conversation with your principal — whether that's a CEO, a department head, a colleague — about how AI changes the priorities of the company and your own role. I had that conversation with my CEO recently, and it was genuinely useful. We aligned on what's next. I'd encourage everyone to do the same.

Ceci [35:50]

That's all for today's episode. Thank you so much for listening to—

Bea [35:54]

Top of the Ops!

Ceci [35:55]

Bye!

Keep Listening

From the Board to the Trenches
Two ex-VCs on why they left the boardroom behind — and what nobody tells you about becoming an operator.