In this episode, we interviewed Husayn Kassai - Founder and CEO of ollo, an AI deployment platform helping organisations build and own AI agents rather than renting a pile of disconnected tools
Husayn previously co-founded Onfido, scaling it to hundreds of people across multiple regions before its $650M acquisition - his current work centres on human-owned AI capability, agent deployment, and building organisations that can use AI without surrendering control
His core argument: AI adoption was never an access problem, it's a deployment problem - and as execution gets cheap and abundant, human judgement, context and ownership become the scarce layer
Key takeaways:
The hard part isn't access to intelligence, it's deployment into complex, permissioned organisations - the model doesn't know your approval thresholds, exceptions, or who actually owns a decision
Deploy workflows, not chatbots - an agent that cites its sources, respects permissions, and refuses to answer on weak evidence is a repeatable support workflow, not a demo
The strategic question isn't which model to buy - it's whether your organisation can absorb frontier capability, and which workflows are ready with clear ownership
Budget shifts from AI tools to AI deployment systems - the layer that connects context, permissions, workflows, evaluation, governance and adoption
🔗 Connect with Husayn
Who are you and what do you do?
I'm Husayn Kassai, founder and CEO of ollo. I previously co-founded Onfido, and now I'm building ollo to help companies move from buying AI tools to building AI capability they can actually deploy, govern, and improve inside the organisation.
What problem did you see that everyone else was missing?
Most people treated AI adoption as an access problem: buy ChatGPT, roll out Copilot, run a prompt workshop and productivity will appear. I think that misses the hard part. The problem isn't access to intelligence; it is deployment into complex, permissioned organisations.
Every company has workflows that live half in systems and half in people's heads. The model doesn't know your approval thresholds, your exceptions, your customer promises, your undocumented workarounds, or who actually owns a decision. As AI execution becomes cheaper and more abundant, human judgement, context and ownership become the scarce layer. That is what we built ollo around: turning tacit organisational know-how into reusable, governed AI capability.
AI doesn't become valuable because it is powerful; it becomes valuable when it is wrapped in the right context, controls and operating model.
Walk us through one concrete way your work changes what companies actually ship - a real workflow, not the abstract
A simple example is onboarding or HR support. Before ollo, an employee asks a question in Slack, someone in HR searches Drive, checks a policy in Notion or Confluence, looks at BambooHR or another system and then replies manually. The same question comes back the next week.
With ollo, we connect the relevant sources, configure permissions properly, and co-build an agent with the HR or People team. The agent can answer in Slack with citations, explain where the answer came from, refuse to answer if the evidence is weak and surface patterns in what people keep asking. The company isn't just deploying a chatbot; it's deploying a repeatable support workflow with evidence, permissions and ownership built in. That changes what teams can safely put into production.
What's the most common thing senior leaders get wrong about AI?
They think the strategic decision is which model to buy. It's not. Models matter, but most companies will have access to broadly similar frontier capability. The real question is: can your organisation absorb that capability?
Senior leaders often ask, "Should we use Claude, GPT, Gemini or Copilot?" The better question is, "Which workflows are ready, who owns them, what context does AI need, what can it touch, what should it never do and how will we know it worked?" AI doesn't become valuable because it is powerful; it becomes valuable when it is wrapped in the right context, controls and operating model. That is where most AI strategies either compound or die.
The company isn't just deploying a chatbot; it's deploying a repeatable support workflow with evidence, permissions and ownership built in.
What's in your AI stack? The one tool you rely on every week?
I use AI for almost everything that benefits from compression: turning long threads into decisions, turning rough thinking into structured memos, comparing options, preparing for meetings and stress-testing messaging. Internally, we use ollo itself for the work we want customers to do: search our own company knowledge, build agents and replace legacy workflows with AI-enabled ones.
The weekly workflow I rely on most is using AI to turn scattered context into a clear operating view: Slack threads, meeting notes, Notion pages, customer conversations and product updates. The value isn't that AI writes a prettier summary; it is that it helps me see what changed, what matters and where judgement is needed.
What does your work actually look like day to day - the real version, not the headline?
The real version is much less glamorous than "building the future of work." It is customer calls, product reviews, messy positioning debates, hiring conversations, writing, rewriting and trying to make sure the company doesn't confuse motion with progress.
A lot of my day is translating between worlds: what customers say, what the product can do, what the team believes, what the market is rewarding and what we need to become. I spend time on the edge cases because that is where the truth usually is. The job isn't to have the best AI opinion in the room; it is to keep forcing the company back to real workflows, real users, real constraints and real proof.
Where is your field in 12 months - one specific prediction?
In the next 12 months, I think we'll see companies become much more disciplined about measuring AI by what is actually deployed, rather than what has been piloted. My specific prediction: the winning category will shift from AI tools to AI deployment systems. Companies will still buy models and copilots, but budget and attention will move toward the layer that connects context, permissions, workflows, evaluation, governance and adoption.
Where should readers find you, and what should they read first?
The best place to find me is on LinkedIn, and the best place to understand ollo is our website: askollo.com. I'd start with anything I've published around human-owned AI capability - the idea that the more powerful AI becomes, the more intentionally human its deployment needs to be.
If you are a leader, I wouldn't start by reading another model comparison. Start by mapping one workflow in your organisation that is painful, repeated, and currently dependent on someone's judgement. Ask what context AI would need, what authority it should have, where it should stop and who would own improving it. That exercise is usually more useful than another AI trends report.
AI Central Voices is where the AI Central team sits down with the founders, executives, and builders shaping AI - going behind the scenes of how they operate, what they're betting on, and where the industry goes next.
Want to be featured, or have an event you'd like us to cover? Reach out at [email protected]
