Lessons from Adrian McDermott
Adrian McDermott has been Zendesk’s CTO since its early days, helping scale the company from a startup into a global platform. He argues that customer service must evolve from a cost center to a driver of brand loyalty, supported by human-centered AI. This profile covers his philosophy on engineering, automation, and leadership.
Part 1: The Evolution of Customer Experience
- On the old model: "Historically, businesses treated support as a cost center, where success was measured by how fast you could get the customer off the phone." — Source: Modern CTO Podcast
- On brand loyalty: "The interactions you have when things go wrong define your relationship with a brand far more than the initial purchase." — Source: Verdict Interview
- On friction: "Customers don't want to explain their problem to three different people. They expect context to travel with them." — Source: Speechless Podcast
- On metrics: "Average handling time is a dangerous metric if it incentivizes agents to prematurely close tickets without actually solving the underlying issue." — Source: First Round Review
- On empathy: "Technology cannot replace empathy, but it can remove the administrative burden so humans have the time to actually be empathetic." — Source: Zendesk Blog
- On proactive support: "The best customer service happens before the customer even realizes they have a problem." — Source: Modern CTO Podcast
- On self-service: "A knowledge base is only as good as its search function. If users can't find the answer in ten seconds, they will open a ticket." — Source: Verdict Interview
- On omni-channel: "Being on every channel isn't the goal. The goal is being on the channels your specific demographic actually uses to communicate." — Source: Speechless Podcast
- On agent experience: "The interface the support agent uses dictates the quality of service they provide. A clunky backend means a frustrated customer." — Source: First Round Review
- On personalization: "Greeting a returning customer by name is baseline. True personalization is knowing what they bought, why it might be failing, and having a fix ready." — Source: Zendesk Blog
Part 2: The Reality of Artificial Intelligence
- On generative models: "Generative AI is inherently creative, but enterprise customer service requires deterministic, reliable answers." — Source: Modern CTO Podcast
- On AI hallucination: "In consumer apps, a hallucination is funny. In financial services or healthcare support, it is a liability." — Source: Verdict Interview
- On implementation: "Don't buy AI for the sake of having AI. Buy it to solve a specific, measurable bottleneck in your routing or resolution process." — Source: Speechless Podcast
- On trust: "The adoption of AI in the enterprise moves exactly at the speed of trust. If the business doesn't trust the model, it won't deploy it." — Source: First Round Review
- On agent assist: "The immediate value of language models is acting as a co-pilot for the human agent, summarizing past interactions in seconds." — Source: Zendesk Blog
- On training data: "An AI model is a reflection of your underlying knowledge base. If your documentation is a mess, your bot will be a mess." — Source: Modern CTO Podcast
- On intent detection: "Understanding what a customer is actually asking, despite typos and frustration, is where machine learning shines." — Source: Verdict Interview
- On cost reduction: "AI will lower the cost of simple transactions to near zero, shifting the budget toward highly trained specialists handling complex edge cases." — Source: Speechless Podcast
- On transparency: "When a customer is talking to a bot, they should know it immediately. Deception destroys brand trust." — Source: Zendesk Blog
- On fine-tuning: "General models are good, but fine-tuning a model on your company's specific historical ticket data is what drives actual resolution rates." — Source: First Round Review
Part 3: Scaling Product and Engineering
- On hiring engineers: "When you interview a thousand engineers, you start to look for curiosity rather than raw coding speed." — Source: First Round Review
- On technical debt: "You have to accept a certain amount of technical debt to find product-market fit, but you must pay it down before trying to scale globally." — Source: Modern CTO Podcast
- On monoliths vs. microservices: "Everyone wants to start with microservices, but a well-structured monolith is often the right choice until your team size forces a split." — Source: Verdict Interview
- On engineering managers: "The transition from writing code to managing people who write code is a complete career change, not a promotion." — Source: First Round Review
- On global teams: "Distributed engineering only works if communication is asynchronous by default and documentation is treated as a first-class citizen." — Source: Speechless Podcast
- On deployment: "If deploying to production is a high-stress event that requires a weekend, your tooling is fundamentally broken." — Source: Modern CTO Podcast
- On product-led growth: "The software should explain itself. If a user needs to read a manual to start a trial, the design has failed." — Source: Verdict Interview
- On system reliability: "Uptime is not a feature; it is the absolute baseline expectation. Nobody praises you for the lights staying on." — Source: First Round Review
- On feature bloat: "Saying no to a customer request is painful, but adding a feature that only a tiny fraction of users need will eventually suffocate the product." — Source: Zendesk Blog
- On refactoring: "Never stop product development entirely for a rewrite. You have to rebuild the engine while the plane is still flying." — Source: Modern CTO Podcast
Part 4: Leadership and Culture
- On evangelism: "A modern CTO cannot sit in a dark room designing architecture. You have to be out talking to customers and explaining the vision." — Source: Verdict Interview
- On failure: "Blameless post-mortems only work if the leadership team actually models vulnerability when they make their own mistakes." — Source: First Round Review
- On company values: "Culture isn't what you write on the wall. It is the behavior you tolerate when things are highly stressful." — Source: Modern CTO Podcast
- On retention: "Engineers leave when they stop learning. If you keep giving them hard problems to solve, they will stay." — Source: Speechless Podcast
- On alignment: "Product and engineering must report to the same North Star metric. If product is measured by features and engineering by stability, they will fight." — Source: Verdict Interview
- On impostor syndrome: "Every leader operating at scale feels like they are making it up as they go. The trick is building a team that catches the things you miss." — Source: First Round Review
- On diversity: "Building software for a global audience requires a team that actually reflects that global audience's perspectives." — Source: Zendesk Blog
- On giving feedback: "Direct feedback is a form of respect. Withholding criticism because it feels uncomfortable is a disservice to your team." — Source: Modern CTO Podcast
- On remote work: "The watercooler moments are gone, so you have to engineer serendipity through intentional, unstructured time in meetings." — Source: Speechless Podcast
Part 5: Automation and Service Debt
- On service debt: "When you implement cheap automation that traps customers in endless loops, you accumulate service debt that eventually destroys loyalty." — Source: Speechless Podcast
- On bad bots: "A chatbot that acts like a glorified FAQ menu and refuses to hand over to a human is worse than having no bot at all." — Source: Modern CTO Podcast
- On human handoffs: "The transition from bot to human must be seamless. The agent needs to see the entire chat transcript immediately." — Source: Verdict Interview
- On automation limits: "Automate the repetitive password resets and order status checks. Leave the complex negotiations and apologies to humans." — Source: First Round Review
- On customer intent: "Automation fails when it forces the user down a rigid decision tree instead of dynamically responding to their actual intent." — Source: Zendesk Blog
- On proactive triggers: "If the system detects an error on a user's screen, automation should proactively offer a solution before the user even complains." — Source: Modern CTO Podcast
- On measuring success: "Don't measure automation by how many tickets it deflected. Measure it by whether the customer's issue was actually resolved on the first try." — Source: Speechless Podcast
- On continuous improvement: "An automated workflow is never finished. You have to monitor where users drop off and constantly adjust the logic." — Source: Verdict Interview
- On IVR systems: "Pressing numbers for support is an outdated model. Modern routing should know who is calling and predict why based on recent activity." — Source: First Round Review
Part 6: Building vs. Buying Software
- On core competencies: "Only build software that is a unique differentiator for your business. Buy absolutely everything else." — Source: Modern CTO Podcast
- On open source: "Using open source accelerates development, but you have to understand the maintenance burden you are taking on." — Source: First Round Review
- On platform ecosystems: "A successful enterprise product eventually becomes a platform. You cannot build every integration yourself; you need an ecosystem." — Source: Verdict Interview
- On evaluating vendors: "When buying software, evaluate the vendor's API first. If the API is an afterthought, the product will be a nightmare to integrate." — Source: Speechless Podcast
- On vendor lock-in: "Lock-in is inevitable at scale. The question is whether the vendor provides enough ongoing value to justify the switching costs." — Source: Zendesk Blog
- On internal tools: "Engineers love building internal tools, but those tools often become unmaintained orphans when the creator leaves the company." — Source: Modern CTO Podcast
- On cloud infrastructure: "The public cloud won the infrastructure war. Spending engineering cycles managing physical servers is a waste of capital." — Source: First Round Review
- On security: "You can outsource your infrastructure, but you cannot outsource your responsibility for customer data security." — Source: Verdict Interview
- On API design: "Treat your public API with the same care as your user interface. For developers, the API is the product." — Source: Speechless Podcast
Part 7: Future of Work and Agentic AI
- On autonomous agents: "We are moving from bots that answer questions to autonomous agents that can actually execute tasks across different systems." — Source: Modern CTO Podcast
- On the agent's role: "The human support agent of the future will function more like an auditor, reviewing the work of AI agents and handling the exceptions." — Source: Zendesk Blog
- On seat-based pricing: "As AI handles more volume, traditional software pricing based on human headcount will have to shift toward outcome-based models." — Source: Verdict Interview
- On skill shifts: "The premium skill in customer service will shift from rapid data entry to complex problem-solving and emotional intelligence." — Source: Speechless Podcast
- On system integration: "Agentic AI is useless if it cannot write to your database. The future is about giving AI the permissions to safely take action." — Source: First Round Review
- On job displacement: "Technology consistently alters the nature of work, but historically it creates new categories of jobs to replace the ones it automated." — Source: Modern CTO Podcast
- On continuous learning: "The half-life of technical skills is shrinking. Engineers must optimize for adaptability rather than mastering a single framework." — Source: Verdict Interview
- On collaboration: "Humans and AI will work in a loop. The AI proposes a solution, the human approves it, and the AI learns from the approval." — Source: Zendesk Blog
- On AI governance: "Allowing autonomous agents to update customer records requires entirely new frameworks for auditing and compliance." — Source: Speechless Podcast
Part 8: The Role of the Modern CTO
- On technical vision: "The CTO's job is not to pick the database. It is to ensure the technical architecture directly supports the business strategy." — Source: First Round Review
- On translating tech: "You must be able to explain complex technical trade-offs to a board of directors using language they actually understand." — Source: Modern CTO Podcast
- On staying close to the code: "You can't write production code as a CTO, but you have to read enough of it to know if your engineering standards are slipping." — Source: Verdict Interview
- On navigating hype: "When a new technology like generative AI arrives, the CTO must separate the marketing noise from the actual engineering utility." — Source: Speechless Podcast
- On organizational design: "Shipping fast is often a function of org design. If a team has too many dependencies, they will ship slowly regardless of their talent." — Source: First Round Review
- On legacy systems: "You can't mandate a rewrite without context. You have to convince the business why untangling legacy code will make them money in the long run." — Source: Zendesk Blog
- On cross-functional alignment: "If the CTO and the Chief Revenue Officer are not talking every week, the company is likely building things that won't sell." — Source: Modern CTO Podcast
- On the CTO lifecycle: "The skills required to be the CTO of a fifty-person startup are entirely different from those needed at a ten-thousand-person public company." — Source: First Round Review
- On intellectual honesty: "The most important trait for a technology leader is the willingness to admit when an architectural bet turned out to be completely wrong." — Source: Verdict Interview