Abe Ankumah's Path Through Four Platform Shifts & What It Reveals About Rewiring Organizations for AI
Engineering organizations are shipping more with AI than ever, and while the dashboards look great, release counts are climbing, demos are multiplying, and every team has a pilot to show. The harder question for a VP of Engineering or CTO now is whether any of that activity has changed how the business actually runs, or whether it has just layered new tools onto old processes and called it transformation.
Abe Ankumah has built through four platform shifts at the infrastructure layer, from the internet and mobility to cloud and AI. He studied electrical engineering at Caltech, joined Fulcrum Microsystems' founding engineering team, and later worked at Aruba Networks and Meraki. In 2013, he co-founded Nyansa, an early AIOps network analytics company that VMware acquired in 2020, and he went on to run VMware's SD-WAN and SASE business as VP and GM. More recently, he served as Chief Product Officer at 1Password, and today he continues to invest, advise, and build alongside founders focused on AI infrastructure and the global ecosystem.
On Episode 15 of TechPod Talks, Stephen Koza, EverOps’ CEO, presses Abe on the first line of his LinkedIn bio, a line that claims the biggest mistakes in AI right now look like progress. The full conversation moves from silicon tape-outs to agent credentials to the change management work that decides whether AI investments pay off. Listen to the full episode and keep reading for a top-down look at how engineering leaders can tell real AI transformation from motion, and which decisions about process, identity, and people really belong on this quarter's agenda.
Most AI Adoption Today Is Filming the Stage Play
Abe names the pattern behind most disappointing AI programs with a word from design history. Skeuomorphism describes the habit of fitting every new technology into the shape of the old one. When cinematography first arrived, he explains, filmmakers pointed the camera at a theater stage and recorded a play. The camera worked perfectly, and the medium stayed exactly as it was until someone took the camera outside and started making movies.
Engineering organizations repeat that pattern with AI every day. A team discovers it can vibe code, so it cranks out more features, more internal tools, and more prototypes. The output is real, and so is the feeling of momentum. Abe argues that the feeling is the trap, because the underlying process, the handoffs, and the decision rights all stay intact.
"If you are adopting AI without changing your process, without actually rethinking things from a first principles perspective, it's going to feel like a lot of progress… but you're actually not going to put yourself in a position to transform as much right now."
For a leader, the test is whether a given AI initiative changed the shape of the work or only its speed. Abe's blunter phrase for speed without a rewired process is "lipstick on a pig," and the data supports him. McKinsey's State of AI in 2026 survey found that nearly three-quarters of AI high performers report fundamentally redesigning workflows, compared with about one-quarter of other respondents. The companies capturing measurable EBIT impact are the ones redrawing the workflow around the capability.
An AI Strategy Expires Faster Than Any Platform Shift Before It
Abe has lived through the internet, mobility, and cloud transitions, and he says the AI shift is moving on a more compressed timeline than any of them. The practical consequence is that a sound AI plan from twelve months ago can be the wrong plan today. Abe points to how quickly the conversation has moved. The industry started with chatbots, and he asks when anyone last heard about a chatbot that genuinely impressed them. The topics now are long-running agents, harnesses, and fine-tuned models, and the topic of the moment keeps turning over from one year to the next.
That pace changes how leaders should govern AI investments. Abe's advice is to keep adopting and keep using the tools, while continually asking how to rebuild your approach around what is possible today. A roadmap built on last year's capabilities locks the organization into last year's workflows, which is the stage-play problem again on a longer timescale.
The decision this creates for a CTO is about review cadence and reversibility. Architecture choices that assume a fixed model capability age badly, so the strongest posture keeps integration layers thin and swappable, revisits use cases on a fixed schedule, and treats every AI workflow as a candidate for redesign when the underlying models improve. A quarterly review that asks what has become possible since the last one is cheap insurance against building an expensive skeuomorph.
Every Agent Inside the Enterprise Is a Privileged User
Abe joined 1Password as Chief Product Officer in 2025 to help the company move from a password manager to an identity security company built for AI. He describes 1Password as a product built on trust, and he says the company held custody of over 1.3 billion credentials while he was there. The challenge his team worked on reframes how engineering leaders should think about agents.
"Every agent within the context of an enterprise is effectively a privileged user. So what does that mean? How do you actually provide credentials just for the task, just in time?"
The implication for platform and security leaders is concrete. An agent with standing access to production systems, customer data, or deployment pipelines carries the same blast radius as an over-provisioned engineer, and it acts faster and at greater volume. Scoped, task-level, just-in-time credentials become a design requirement for any agent headed to production. Stephen describes connecting 1Password to his own terminal so a coding agent could pull an SSH key on demand, and he calls it one of the cooler moments of his week. Agents are becoming the users, and systems are going headless.
This is where many agent programs stall. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Identity and access design for agents is one of the most direct ways to close the risk-control gap before an initiative reaches production.
Durable AI Starts With Designing for Failure Before It Happens
Abe's first startup lesson came from hardware. At Fulcrum, he helped build low-latency switching silicon, and he says that work taught him to own the whole problem, because no one hands a founding team a spec and the market gives feedback fast.
"Silicon is very unforgiving. You can't patch a tape-out. When you tape out a chip, and it comes back, that's what you've got. And so you learn by designing for failure before it happens."
He has carried that instinct through his career, and it applies directly to AI infrastructure. Agents are easy to spin up and expensive to recover when they fail in production, and the failure modes usually trace back to the foundation underneath them. Stephen sees the same pattern across EverOps’ clients. Advanced technology is hard to build and use on top of heavy tech debt and disparate systems. Weak observability, security gaps, and a missing governance model compound the problem. That foundational work, which EverOps has done for 14 years, became far more critical once AI entered the picture.
For engineering leaders, designing for failure means deciding what an agent can break before it ships. That covers the observability needed to see what an agent did, the governance that decides who owns its output, and the rollback path when it goes wrong. Abe also notes that low-latency data movement, the problem he worked on at Fulcrum, has become even more urgent as AI infrastructure scales.
Customer-Backwards Thinking Shows Which Processes to Rebuild First
The skeuomorphism trap raises an obvious question for leaders: where do you begin rewiring? Abe's answer comes from Dominic Orr, the Aruba CEO who hired him out of business school as chief of staff and promised to teach him everything he knew about running a public company. Abe says Orr delivered in spades, and the lesson that stuck was what Orr called "brutal intellectual honesty."
Orr thought from first principles and from the customer backward. He studied what was changing in the customer's world, then challenged the team to ask, given those facts, "what has to be true for what we're doing to actually endure." Abe's example is the iPhone and iPad. Aruba sold enterprise wireless before those devices shipped, and the day they arrived, universities and enterprises realized wireless had to be secure and pervasive across the whole organization. Customers still hesitated to cut the cord, so Aruba built the product around that hesitation, addressing every obstacle directly, through partners, and through education.
Abe applies the same discipline when he advises leaders on AI today. He tells them the real question goes deeper than whether the organization uses AI. Leaders should ask which processes or activities have become irrelevant, or could be fundamentally rethought with these capabilities. That framing gives a VP of Engineering a practical way to prioritize. Start with the workflows where the customer experience would change most, define what must be true for the new version to endure, and rebuild those first.
Change Management Is the Binding Constraint on AI in Established Companies
The section a CTO acts on first is the one Abe calls the real constraint. AI, he argues, is more than a technology because it rewires everything around it. That makes it as much a change management challenge as a technology and adoption challenge. People look at AI and ask what it means for their jobs and livelihoods, and leaders have to answer that question as part of the rollout.
"If you're a new organization, you can be AI native from day one. But if you're an established organization, that change has to be very deliberate."
Stephen sees that constraint in nearly every engagement. Very few leaders can turn a public company upside down overnight, and the bigger the company, the longer new ways of working take to settle in. The job, as he describes it, is helping an organization change fast enough to capture the benefits and slowly enough for people to absorb the change. That surfaces policy questions most AI plans skip. When a team deploys an agent for a specific use case, is using it mandatory or optional? Stephen recalls a HubSpot VP who rolled out Copilot as optional a few years ago and tracked adoption, finding some engineers strongly opposed and others fully committed.
Abe also points to a signal from the frontier AI labs themselves. In recent months, they have stood up services businesses to help enterprises adopt AI, and Stephen reads that as an implicit admission that deploying agents in the enterprise is harder than anyone expected. Abe connects the trend to the forward-deployed engineering model, which Stephen notes EverOps has run for years (TechPods), which are small teams of senior engineers embedded inside a client's systems. Getting agents into production and moving earnings per share takes a solid foundation, a deliberate change plan, and people working inside the business alongside the teams who have to adopt it.
How to Apply These Frameworks This Quarter
- Audit your AI portfolio for skeuomorphs: List every active AI initiative and mark whether it changed the workflow or only accelerated the existing one. Redirect investment toward the initiatives that redrew the process.
- Set a quarterly capability review: Schedule a standing review that asks what has become possible since the last one, and flag any AI workflow built on assumptions the newest models have already overtaken.
- Treat every agent as a privileged identity: Inventory the credentials your agents hold today, then move production agents to scoped, task-level, just-in-time access with an owner for each.
- Write the failure plan before the agent ships: Require observability, a named owner for agent output, and a tested rollback path as entry criteria for any agent moving to production.
- Publish your adoption policy: Decide which AI tools and agents are mandatory and which are optional, share how you will measure adoption, and give managers a clear answer for teams asking what the change means for their roles.
Two recent EverOps posts extend this argument. The infrastructure gap behind stalled AI ROI shows how every deployed agent adds governance, orchestration, and security work. Plus, a look at the 5% getting returns traces the data and infrastructure foundations that separate AI programs that pay off. Both are worth a read if interested in diving deeper into the topics explored throughout this week’s episode.Â
Rewire the Work With a Partner Already Inside It
Abe's warning is that AI adoption without process change produces shiny output and little transformation. Acting on that means an organizational leader has to choose which workflows to rebuild and secure agents as privileged users. It also means hardening the foundation those agents run on and leading the change management that brings teams along.Â
Not sure where to start? The AI Opportunity Assessment helps organizations identify the workflows worth redesigning first, with prioritized use cases and ROI projections. From there, the AI Quick Start delivers a working use case with measurable business impact in eight weeks, and the AI Adoption TechPod embeds a forward-deployed AI engineering team to scale adoption over six months or more.
Ready to turn AI activity into a rewired operating model? Reach out to the EverOps team.
Keep up with TechPod Talks
TechPod Talks brings candid conversations with the engineers and technical leaders building platforms, running cloud, and scaling infrastructure. Subscribe on Apple Podcasts, Spotify, YouTube, or the EverOps podcast page. Connect with Abe Ankumah on LinkedIn and follow EverOps on LinkedIn for clips and episode announcements.
Frequently Asked Questions
What is TechPod Talks?
TechPod Talks is the EverOps podcast for engineers and technical leaders building platforms, running cloud, and scaling infrastructure. Host Stephen Koza, CEO of EverOps, sits down with leaders doing the work to share what is working right now. Episodes are available on Apple Podcasts, Spotify, YouTube, and the EverOps podcast page.
What topics does Episode 15 cover?
Episode 15 covers skeuomorphism in AI adoption, the identity challenge of treating every enterprise agent as a privileged user, and designing for failure from the silicon layer up. Abe Ankumah also discusses customer-backward thinking learned at Aruba Networks, change management as the binding constraint on AI in established organizations, and his work in the African entrepreneurial ecosystem.
Who is Abe Ankumah?
Abe Ankumah is an infrastructure and product leader who co-founded Nyansa, an AIOps network analytics company acquired by VMware in 2020. He later ran VMware's SD-WAN and SASE business as VP and GM and served as Chief Product Officer at 1Password. He also serves on the Board of Trustees of the Ashesi University Foundation and now invests, advises, and builds.
What is skeuomorphism in AI adoption?
Skeuomorphism in AI adoption is the habit of applying AI inside existing processes without redesigning them. Abe Ankumah compares it to early filmmakers recording stage plays with a camera. The result feels like progress and produces more output, while the workflow itself stays unchanged.
Why should AI agents be treated as privileged users?
Agents inside an enterprise often hold access to sensitive systems and act at machine speed, so Abe Ankumah argues each one is effectively a privileged user. That calls for scoped credentials provided just in time for a single task, with clear ownership and auditability.
How does this episode connect to EverOps' work?
The episode's focus on rewiring processes and strengthening foundations maps directly to how EverOps embeds TechPods inside client organizations. The AI Opportunity Assessment and AI Adoption TechPod help leaders choose and scale the right AI workflows, and the Observability practice builds the visibility agents need in production.
