The spirit and the letter: running compliance at startup speed
A conversation about turning compliance frameworks into everyday discipline, working with auditors instead of against them, and what AI changes about accountability.
- HostGreg Gzik
- Founder & Principal, Autom8
- GuestAaron Verones
- Former Head of Engineering, Wrk and Catio
Aaron Verones has run compliance programs at two startups: Wrk, a Montreal automation company of up to 110 people, and Catio, a 12-person AI company. Between them he has taken teams through SOC 2, HIPAA and Canada’s PIPEDA. In this conversation he and Greg Gzik compare the startup and enterprise views of compliance: how to make frameworks readable, how to prove a control works, how to keep a program alive between audits, and why AI-assisted engineering raises the bar on accountability rather than lowering it.
What leaders can take from this
- 01Compliance is a leadership problem before it is a technical one. Give every control an owner, and follow up.No more annual fire drills
- 02Read each requirement for its spirit. When the letter doesn’t fit how you build, make the case to your auditor yourself.Strong and weak compliance officers
- 03Interview auditors the way you would a hire, and treat them as a source of clarity, not an adversary.Advice for first-timers
- 04You can’t backfill evidence. Surface a gap early, document it and fix it; a red status is information, not failure.Evidence, and what to do about gaps
- 05Change your architecture as often as you need to. Keep a paper trail and an accountability trail while you do.New systems mid-audit
- 06AI raises the bar on accountability: engineers own what their agents ship, and attackers have the same tools.AI and accountability
Edited for clarity. Filler words, false starts and brief acknowledgements have been removed and some sentences lightly reworded for readability; the substance is unchanged. The video’s opening introduction is not transcribed.
Introductions
Thank you so much for joining me, Aaron. I know you quite well, but my audience probably doesn’t. Why don’t you quickly introduce yourself and your background?
For sure. My name is Aaron. I’m a software engineer by trade, with a background in distributed systems, high-performance computing, big-data knowledge stores, all that fun stuff. At the dawn of AI, I jumped ship to an AI lab in the States to figure out what the deal is with AI nowadays.
My roles have mostly been in engineering leadership. I was head of engineering at Wrk, an automation company based in Montreal, and most recently head of engineering at Catio, an AI company based in the Valley.
One of the interesting things about your experience compared to mine is that I’ve spent most of my career in similar engineering leadership positions, but in an enterprise context. What was the size of the companies you were contributing to?
Smaller, for sure. Wrk was 110 people at its biggest; Catio was 12. I’ve definitely seen the differences and similarities between our work. It’s fun when we’re solving similar problems at completely different scales, and it’s also interesting to see the unique problems you face that I don’t, and vice versa.
Getting started in compliance
You’re no stranger to the complex world of compliance. Tell me how you got started in this space and what kinds of compliance frameworks you were working with.
I’ve run compliance programs at two different companies, and they’re interesting examples because they were at different stages of maturity and had different reasons for pursuing compliance.
The first was Wrk. I was leading a team of 17 engineers at that point, and we did SOC 2 Type 2 and HIPAA. We also needed to comply with PIPEDA, Canada’s federal privacy legislation. PIPEDA is a little different because it’s legislation rather than an assurance framework like SOC 2, but from an operational perspective a lot of the work was similar: translating requirements into policies, establishing procedures for handling customer data, all that kind of stuff.
At Wrk, our compliance program was led by our COO and our compliance officer, and I was responsible for the engineering side: essentially every control, policy or procedure that touched our production systems or handled customer data. Mostly, I took inventory of our whole engineering environment. We needed to understand how customer data flowed through our systems, where it was stored, and what operational processes our customers depended on. Once we understood that surface area, we could take the compliance controls, figure out how they applied to it, and implement them.
Making frameworks readable
“You have a tree of accountability, and software like Vanta lets you see it all at a glance and target the controls that need fixing.”
Let’s pause there. One of the challenges I’ve seen, and I think you put it well, is that your compliance burden is not necessarily any one standard or framework. It can be a mixture of legal compliance, specific commitments you’ve made in contracts, and a particular framework you intend to target, like SOC 2 Type 2 or ISO 27001, because you’re trying to get into a new market.
These frameworks are not exactly good reads. I have no intention of printing out the SOC 2 Type 2 standard and curling up by the fireplace on a Friday night to read it. I’m also not a lawyer by training, and neither are you. So how do you interpret these requirements, which can often number in the thousands? I think Adobe had something like 4,200 requirements from various frameworks that it had to satisfy. How do you turn that into something sensible?
The closest answer is: you automate. You use software like Drata or Vanta. They come with these compliance frameworks prepackaged in a way that’s accessible to non-legal, technical readers, and they have a whole suite of automated tools as well.
SOC 2 is a good example. There are five trust services criteria in SOC 2, and as a business leader you simply say which ones are relevant to you. You don’t need a background in compliance or in law. Out of security, availability, processing integrity, confidentiality and privacy (I’m glad I remembered those) you just choose the ones that are relevant to your business. At Wrk we did security. At Catio, we also did availability and confidentiality.
Then we used Vanta. Instead of a thousand pages of documentation to read, Vanta segments everything into a tree of controls. Each control has a suite of automated tests, or documents you need to upload, and you can assign each one to the leader in your organization who is accountable for it. In a larger organization, it’s not going to be one person responsible for getting everything done. You have a tree of accountability, and software like Vanta lets you see it all at a glance and target the controls that need fixing.
When a control doesn’t fit your architecture
“We needed to show the auditors both sides: what Wrk was responsible for and what Google was responsible for.”
That’s great. The category these systems fall into is governance, risk and compliance: GRC. Fun fact: I started at a GRC company, then called Galvanize and now called Diligent. So if anybody using the Compliance Maps application finds a bug, don’t come to me. I promise my code was perfect.
Even with these programs, there’s a very difficult task in turning legalese into something actionable, and what’s actionable for one organization isn’t necessarily actionable for yours. Have you had a situation where, even with a simplified GRC program, you still had to do additional interpretation to make a requirement make sense at Wrk or Catio?
For sure, and there’s a great example from Wrk involving managed cloud infrastructure. Traditional infrastructure security procedures assume that a company administers its own servers. If you manage a virtual machine, it’s pretty straightforward to interpret things like operating-system security patches or server hardening. But that wasn’t our architecture. Wrk primarily operated on Google Kubernetes Engine: we deployed Docker containers containing application code. We didn’t manage our servers; Google did.
So when an auditor asked us to demonstrate how we patched the operating systems of our underlying servers, we didn’t administer those machines. We needed to explain the shared responsibility model. Those controls are operated by Google Cloud, and Google has its own SOC 2 reports and vendor agreements. We had our own responsibilities, like upgrading to the latest GKE versions, but we needed to show the auditors both sides: what Wrk was responsible for and what Google was responsible for.
Our auditors had never seen this before. They came from a traditional background of people managing their own servers, and we had to explain to them what serverless technologies were. It’s one of the things I learned at Wrk that informed how I selected auditors at Catio. I ran the whole compliance program at Catio, so I got to choose the auditors, and I was definitely interviewing to see whether they’d seen serverless before.
Strong and weak compliance officers
“A strong compliance officer works with engineering, or whoever else the control applies to, and finds the spirit of the law.”
I think this is a really useful tension, and I’ve had similar experiences. There’s a real benefit in using a managed vendor, like a cloud provider, to handle a lot of your security. You mentioned the shared responsibility model: security of the cloud versus security in the cloud. The provider makes sure the cloud itself is secure, but you still have to be secure within your own cloud context, implementing authentication, authorization and everything else we know as application developers.
Sometimes you see companies whose internal auditors or InfoSec teams come from a traditional IT security background, which is great because they bring a wealth of experience. But new technologies like serverless might be unfamiliar to them, and there can be a pull to say, “Maybe we shouldn’t adopt that, because it might put our compliance at risk.” Do you have experience with that, and how do you manage that kind of conflict?
If you zoom out, there’s a difference between a strong compliance officer and a weak one. A weak compliance officer looks at the letter of the law, says “the framework says this,” and tosses it over the fence to engineering, no matter how much additional work that creates or how it affects the unit economics of the business. It’s a very unaccountable way of working.
A strong compliance officer works with engineering, or whoever else the control applies to, and finds the spirit of the law. How do we operate faithfully in accordance with the spirit of this requirement, so that we’re always available for our customers, we have strong business continuity, and we’re good stewards of their data? Maybe there’s a version that’s easy for the company to do and satisfies 80% of the letter of the law. Then I, as the compliance officer, go and make the case to the auditors. I’m picking up some of the burden. A strong compliance officer navigates that tension well.
One control, end to end
“You’re often responsible for the effectiveness of controls that are operated by subject-matter experts outside your team.”
Well said. Earlier you mentioned the importance of taking requirements from these frameworks, often several of them, and turning them into a smaller set of controls that dictate how your organization will satisfy them. That gives you a much smaller surface area to think about. Could you take one control, maybe your favorite, and walk us through the process: developing it, wording it appropriately, implementing it, and everything in between?
I don’t know about a favorite, but I’ll talk about one that isn’t an engineering control, because engineering controls are easy for me. As head of engineering, I own the control and the implementation, it’s validated by an automated test, and the auditor sees it at audit time.
A trickier one, which we implemented at Catio, was our customer support control. SOC 2 requires you to maintain an external-facing customer support mechanism, and we implemented it through several channels. We introduced Intercom, which gave customers a direct support interface in our app. We had dedicated Slack or Microsoft Teams channels with individual customers, so we could reach them directly in the tools they were already using. And we documented our support channels on our website, so somebody researching Catio for the first time could find our support information. Even if they asked ChatGPT how to get support at Catio, it could read our public documentation and tell them who to email.
But the actual operation of those channels belonged to our customer success team. I was responsible for the overall compliance program and ultimately accountable for making sure this control kept working, but it was a cross-team dependency. So I assigned control owners in Vanta, explained to the responsible team what the control required, and periodically reviewed whether it was being followed. I needed to give the customer success team enough information to own the process themselves. I didn’t want to just quietly do it in the background.
That was an important lesson in compliance leadership. You’re often responsible for the effectiveness of controls that are operated by subject-matter experts outside your team. So you do what any leader does: communicate your expectations, establish accountability, and be disciplined about following up.
Evidence, and what to do about gaps
“You can’t truthfully produce evidence that those communications happened in January if they didn’t.”
Brilliant. And you’re not the only one who follows up; auditors do too. An auditor’s job is really interesting, because it’s in the company’s own interest to put the best possible case for its compliance maturity forward, and sometimes that involves some creative interpretation of language. As an auditor, I don’t want to stand for that. I want to make sure the companies I’m auditing really are conforming not just to the letter of the law but to its spirit as well. So what I need is evidence that you’re actually operating the control you say you are. How do you show an auditor that evidence?
For automated tests, it’s easy. There are integrations with your AWS environment, or wherever you run, that establish a lot of the proof, like the fact that all your disks are encrypted. AWS reports it, and it gets pulled into your GRC software.
Others, like the customer support control, have to be demonstrated manually. We ended up taking screenshots of customer interactions to prove that customers really were requesting support from us, and giving those to the auditors along with documentation of our support channels: look, real customers are actually doing this.
An interesting story from Wrk: sometimes you have gaps in your evidence, and both you and your auditors will see them. At Wrk we were HIPAA compliant, and one of HIPAA’s requirements is that you communicate security-related changes internally. Somehow that was missed in the first audit period. It’s April, or whenever, and we realize we haven’t been sending these required communications for January, February and March.
So you start by communicating all the relevant information alongside the April update, and you put in a recurring procedure so it happens every month going forward. But you can’t go back in time. You can’t truthfully produce evidence that those communications happened in January if they didn’t. So we had a gap. We went to the auditors and explained ourselves: we documented the gap, remediated it, and explained why it happened. The auditors then have to use their judgment to evaluate the significance of that departure. In this case it wasn’t considered material enough to require an exception, but some things are. Sometimes you send a gap to your auditors and they say, “This is going to result in an exception on your report.” So you have to be consistently diligent throughout the year.
No more annual fire drills
“The way you work by default as an organization should be compliant.”
What I’ve found, too, is that a really good external auditor often wants you to succeed. They genuinely care about your maturity. So if you present evidence that you weren’t fully satisfying a control partway through, but you identified it, corrected it and were forthright about it, that often earns you more leeway than trying to hide or cover it up.
I’ve often seen auditors finish a process with no exceptions but a list of recommendations (hint, hint, nudge, nudge) that they’ll be looking for at the next audit. Audit time is maybe a once-a-year or twice-a-year event, and because it’s so infrequent it can turn into an annual fire drill. How do you make sure the compliance program you’ve built, with all its controls and evidence, stays easy to maintain between audit periods? Or have you ever designed something in a way that made it harder to maintain over time, and had to correct it?
It’s a good question. Zooming out again, the message I’d like compliance and engineering professionals to take away is that the technical challenge of implementing these frameworks is not the biggest problem. The biggest problem is making sure accountabilities are spread well across the organization, having the discipline to keep yourself and your teams honest, having the discipline to follow up, and building compliance into your processes. The way you work by default as an organization should be compliant. It shouldn’t be that you work normally and then, at audit time, there’s a fire drill: “Oh no, did we work in a compliant manner? Do we need to retroactively put together justifications for the auditors?” That’s not a good way to work.
Like I mentioned, the number one thing I’d recommend is GRC software like Vanta or Drata. The amount of time these platforms save the organization and the auditors is enormous. Without them you’re doing one of two things: everything manually, or building spreadsheets, scripts and dashboards that reproduce the same functionality anyway.
Wrk had two frameworks, SOC 2 and HIPAA; Catio had SOC 2; and both used Vanta. We wrote all our policies to be compatible with how our business actually worked. I had a dashboard of hundreds and hundreds of controls, with an owner assigned to every single one, and automated messages that would email them if a control was ever overdue, lacking evidence or failing. And I set the expectation with my leaders and delegates that these would be followed up on. If you don’t let things fall through the cracks, they don’t.
I really like that word, discipline, especially because in your example it came with real consequences on an automated schedule. I wish I had the discipline to go to the gym as much as I should, but I don’t. The only way I go is by building it into my work schedule, so there’s a constant reminder that yes, I should be prioritizing this. I love that framework.
And I love what you said a while ago. You have a great phrase: “success is inevitable.” Structure your life, your work and your internal procedures so that success is the path of least resistance. Then you’re not worrying about it. It’s just inevitable.
When the business changes
“The purpose isn’t specifically to have a second human click an approval button. The purpose is to make sure software changes receive appropriate scrutiny.”
So you’ve got your compliance program set up. Your audit isn’t an annual fire drill, your auditor is passing you and you’re not getting exceptions. But your business changes, especially in a startup. The number of times I’ve heard the word “pivot” from a startup founder is too many to count. How do you make sure your controls stay relevant as your business and your technology change?
The biggest thing that comes to mind is AI-assisted coding. In traditional software engineering, a developer writes code, another human developer reviews it, and it’s released to production. So most compliance frameworks, including SOC 2, say you require review before changes. That isn’t how AI-assisted coding is done economically within a business, and we found that out pretty quickly at Catio.
AI-assisted development let our engineers produce code much, much faster. But if you require another human engineer to inspect every change, your review capacity becomes the bottleneck rather than your development capacity. So we started working out what our code review and deployment controls were actually supposed to accomplish. Again, letter of the law versus spirit of the law. The letter of the law says all changes must be reviewed. The spirit of the law is that your systems are maintained by accountable humans who carry liability, that you’re always there for your customers, and that your software development lifecycle is structured to make that possible. The purpose isn’t specifically to have a second human click an approval button. The purpose is to make sure software changes receive appropriate scrutiny.
So we began using AI-assisted code review. AI reviewers with fresh context inspected proposed changes; they had access to our compliance frameworks and all our internal policies, and could point out security problems or other engineering concerns. We also maintained a strict authorization boundary for production. Our AI agents were shipping changes to dev all the time, but as soon as a change had to go to production, it was gated by human review.
That was a technical decision and a business decision that worked for us. But we also needed to talk to our auditors about it. So again, we went to them and said: here’s the spirit of the law, here’s the letter of the law, here’s what’s happening in the world. Can we have the flexibility in our compliance program to show that we’re faithfully executing the spirit of this framework? At Catio, our auditors said yes. After some deep dives into what we were doing, they said we were faithfully following the spirit of SOC 2. Check mark.
New systems mid-audit
“All you need is a paper trail and an accountability trail.”
Fantastic. That’s a great example, because it points to something I definitely struggled with in my previous role: how much should you re-examine a control, versus disagree and commit and move forward?
Here’s one example. We were introducing several new architectures, and we had worded our controls around the existing technology stack. The company’s interpretation of SOC 2 was that we couldn’t release the new version of our software to customers until it had gone through a six-month operating period, which is common in SOC 2 programs. But imagine the disincentive that interpretation creates. It basically means you can’t introduce new product lines to your customers, nearly all of whom require SOC 2, until you’ve been operating for six months. If you’re a startup, your company could be dead in six months. Do you have any experience there?
This goes back to what I said about a compliance officer who says “maximum safety, letter of the law,” throws it over to engineering and takes no accountability. The reality is that I’ve been in startups my whole career. Our systems change constantly; we’re replacing major parts of the architecture on a weekly basis. Your compliance program isn’t meant to prevent you from doing that, and it shouldn’t. It’s there to make sure you’re an effective vendor for your customers, stewarding customer data and maintaining business continuity. None of that is incompatible with replacing major parts of your architecture. All you need is a paper trail and an accountability trail.
Say you’re in a 12-month SOC 2 observation period, as we were, and you expect to introduce new services and infrastructure during it. Those systems won’t have every production control in place on their first day of development, and your auditors should understand that. Again, select for this when you interview them. April 1, we begin building a new system. June 1, it enters a closed alpha. July 1, we complete the production controls and compliance checks. July 2, it begins processing customer data. If you can demonstrate those dates and the applicable system boundaries, with an evidence trail for each stage, give that to your auditors and they’ll be completely fine with it.
The problem is when you say, “This system has been processing customer data since the beginning of the year, but we don’t have evidence that the controls were being followed from the moment it started.” Then your auditors have a justifiable case to have a problem with that.
Maturity is not a checkbox
“Don’t confuse maturity with compliance status.”
I do want to take the opportunity to defend InfoSec departments. Their job is really hard, because engineers want to move fast. Some even say they want to move fast and break things. Well, security is not something you can or should be breaking.
I think this tension often arises because InfoSec teams feel the only way they can influence the roadmap and get the business’s resources is by using the language of security and compliance, as if the roadmap only changes when there’s a security incident or the compliance program is at risk. In my opinion, a healthier relationship is to have a proper security roadmap, so you decouple minimum viable compliance from the idea that it’s the only security or compliance work you do. Don’t confuse maturity with compliance status. You want both, but maturity is a process. It happens over time. It’s not a single checkbox on a row in an Excel spreadsheet.
Absolutely. It goes back to the spirit of the law. A major part of the art of compliance is balancing two objectives that in a perfect world would be completely aligned, but frequently aren’t. There’s the letter of the law: the formal criteria, procedures, evidence obligations and security requirements you need to satisfy to obtain your report. And there’s the spirit of the requirements: the things that actually make your software more secure, your organization more reliable and your customers’ information better protected. A good compliance officer is constantly trying to maximize the overlap between those two. You want the activities that satisfy your auditors to be the same activities that genuinely improve the way your company operates.
Well said. In my experience that’s often a negotiation between engineering and InfoSec, and it’s rarely a case of one party being right and the other wrong. You need both perspectives, and sometimes getting there involves some conflict. From your engineering perspective, have you ever had to negotiate or challenge a compliance process to make sure it achieved what the business really needed to move forward?
I’ll go back to change management, because it’s such a good example. Following the letter of the law on change management, and not being able to use the benefits of AI-assisted development, would have been a catastrophic, maybe even existential, risk to the business. But we didn’t want to abandon change management. The point isn’t to undo everything compliance frameworks and change-management practice have built up as a body of knowledge over the years. The point was to change how we implemented it, because the traditional workflow no longer made economic sense.
Again, it’s about the overlap: you want compliance to work for the business. Being reliable for your customers is business value. Earning your customers’ trust makes you more competitive in the market. It’s a feature.
AI and accountability
“I don’t care what tools you use. I care that you apply them in a way that delivers gold-standard engineering I can stand behind.”
Absolutely. You brought up AI-assisted development, and I even hesitate to use that phrase. More and more, this is agentic development. Yes, humans are still in the loop, because at least as of this recording AI doesn’t quite have the judgment or the taste to deliver something valuable without one. But it’s taking on more and more of the day-to-day of development and of running a business. You mentioned AI-assisted reviews and AI-driven change management. What are some of the other ways you’re seeing AI interact with your compliance programs?
Compliance, again, is about business continuity, safety and being a good steward, and companies need to use AI in a way that makes them better at that, not worse. We talked about AI’s impact on something small like code review controls, but there’s a broader theme of accountability I’ve had to address with all my teams.
When AI-assisted coding first became popular, and now with full agentic development, I noticed a concerning pattern. Developers would vibe-code a piece of software while experimenting with these new tools and ship it to production. Then, when something broke or the code turned out to be non-compliant or insecure, they’d say, “Oh, sorry, that was the AI agent.” It was almost as if involving AI had somehow transferred responsibility away from the engineer.
My message was essentially this: if all I wanted was code production without accountability, I could start ten AI coding sessions myself and have them generate software all day. That’s not what I hire professional engineers for. I hire professional engineers to take responsibility. Your job is to deliver software that works, demonstrate that it satisfies our requirements, and stand behind it as the credible subject-matter expert responsible for that change. And that software needs to be maintainable, deployable and secure for the long term, long after you’ve left the company. I don’t care what tools you use. I care that you apply them in a way that delivers gold-standard engineering I can stand behind.
So well said. That accountability is so critical now. Before, if you wanted to implement a significant feature, you needed a team of engineers and it might take a quarter, two quarters, sometimes a year, depending on its size and scope. Now a single engineer can build a feature. A single engineer can build an entire microservice. At the rate these technologies are going, a single engineer could build an entire startup. So the thing compliance encourages you to do, which is take ownership of what you build, needs to be present no matter how far AI-assisted development or agentic engineering takes us.
AI on both sides of security
“Security by obscurity is going to stop working.”
And by the way, it’s not a zero-sum game. AI and agentic development have made my compliance processes better. Something we had a notorious problem with at Wrk was that the company grew very broadly, very quickly. We had a huge surface area, and it was very hard to patch security vulnerabilities. The time it would have taken to have my team patch every single one would have been uneconomical.
Not so with AI coding. As long as you have a good test suite, you can have AI patch a vulnerability, work out a set of third-party package versions that are compatible, run that through your tests and, once the appropriate change management procedures are applied, release it to production with all your vulnerabilities patched. Our systems are more compliant and more secure than they were before AI development, and that’s how it should be going forward.
Definitely. And the reverse is also true: it’s now the minimum standard of engineering. Just as you can use AI to accelerate your own compliance and security, bad actors can use AI to close the gap between finding a vulnerability in your software and exploiting it. The place we need to get to is machine-to-machine security and compliance. There’s no other path forward, and we can only do that through AI.
I completely agree. The dirty little secret of security on the web is that there are vulnerabilities everywhere. It’s just that not enough of them are valuable to exploit, because of the calculus of the work it takes to exploit them versus what an attacker gains. AI completely changes that calculus. One bad actor can spin up agents with cheap tokens to exploit every vulnerability they can find, faster and more broadly than ever before. The idea that you can live with vulnerabilities as long as they’re not valuable enough to exploit is really going to have to change.
Absolutely. I remember reading an Anthropic article shortly after the Mythos model came out, about an experiment targeting open-source software, some of which had been public for decades. With a little prompt engineering, they were able to use the model to find real, high-criticality exploits in software that had been robustly pen-tested for decades. If these models can do that to mature open-source software, imagine what they can do to a struggling startup that has been focused primarily on features and may not have security front and center.
Yes, and security by obscurity is going to stop working. AI can be so exhaustive that obscurity means nothing. You have to make your systems more bulletproof and airtight, and stop relying on obscurity.
Advice for first-timers
“Don’t be afraid of your auditor.”
Well said. I have a small but mighty audience that includes people from my background in the enterprise and from yours in startups, and some of them are navigating compliance for the first time. Thinking back to when you first got into this, what do you wish you had done differently, or had access to, that would have made your compliance journey easier?
I had a misconception that made my compliance journey harder, one I had to unlearn as I took part in compliance exercises and eventually became responsible for running them. So my advice is: don’t be afraid of your auditor.
A lot of people approach audits as though the auditor is an adversary whose job is to catch them doing something wrong. That’s not a productive way to approach the relationship. Auditors need to maintain their independence, so they won’t operate your compliance program for you, but they’re also an important source of clarity about how your controls will be evaluated.
When you have the opportunity to choose an auditor, interview them carefully, the way you’d interview an employee. Make sure they understand your technology stack and are familiar with the kind of business you operate, and that you can have constructive conversations with them about ambiguous requirements, unusual situations or black-swan events. Whenever something happens that might call your compliance into question, bring it up; you should feel comfortable doing that with your auditor. If you’re unsure whether your procedures or implementation satisfy a control or framework, ask. If you discover a gap, close it. It’s so much better to deal with these issues while they’re small than to hide them or wait until audit time. A collaborative relationship with your auditor is far more productive than an adversarial one.
That’s great advice. I’d add, especially if you’re a technical leader: reach out into your network and talk to real practitioners about how they implement their controls. Your auditor or InfoSec team will give you an interpretation, and given their experience it’s likely to be correct. But if you compare it with your peers’ experience, you may find an approach that not only improves your maturity but is also a much more natural engineering fit for that control than you might otherwise have found.
Discipline, integrity and red statuses
“Compliance is much more an exercise in discipline than in technical complexity.”
Absolutely. And for technical leaders without experience in this who are worried (“I don’t know the law, I don’t understand compliance, and if I get something wrong I can’t just fix it like a code bug; it could really affect my business”), it comes back to what I said earlier. Compliance is much more an exercise in discipline than in technical complexity.
Most professional organizations can figure out how to enable encryption, implement access management and establish change-management procedures. The challenge with compliance is leadership; it’s organizational. How does accountability flow from you, the compliance officer or engineering leader, to the rest of the company? How do you make sure every control has an owner? How do you communicate expectations in a way people understand? And more broadly, how do you build a culture where people act with integrity even when acknowledging a mistake is uncomfortable?
The purpose of compliance is to make the company worthy of the trust its customers place in it. To the leaders out there: if you lead it well, you should be able to look your auditors and your customers in the eye and say with confidence that your company follows both the spirit and the letter of its commitments. To me, that’s what a successful compliance program looks like.
That is brilliantly said. For me, the part that resonated most is having the courage to admit your mistakes, and not treating a green dashboard status as the goal in itself.
About a year and a half ago, I really wasn’t watching what I was eating. I was probably drinking too many beers, and I was overweight. The way I got back to a healthy weight was by tracking it: tracking my calories, what I ate, and reporting my weight faithfully. If I had just reported a green status, it would never have driven me to change my behavior or to end up healthy. So the goal isn’t to make sure your compliance dashboards are all green. In fact, we should celebrate the amber, yellow and red statuses, because they’re an opportunity to genuinely improve our maturity, rather than chasing vanity metrics to hit an arbitrary business target.
Aaron, this has been so much fun. If anyone wants to learn more about who you are and what you’re building right now, where can they find you?
You can find me on LinkedIn, Aaron Verones, or email me at aaron@verones.io. I try to keep it easy. In general, I’m watching for small companies that are heavily adopting AI and putting products into the world that make companies run fundamentally safer, more effectively and more economically. That’s what I’m passionate about. If you’re working at one of those companies, let’s talk.
Awesome. Thank you so much for your time, Aaron. I really appreciate it.
Facing a compliance decision of your own? Bring me the messy version.
Scope, control design, an audit you’re not sure how to approach: think it through out loud with us before it becomes another release bottleneck. No polished brief. No obligation to turn it into a project.