On May 26, we brought a room of senior enterprise leaders together at Toronto Tech Week for something they promised upfront would not be a product pitch or a technical deep dive.
It was a full day: 90 minutes in the morning, a leadership lunch, and 90 minutes in the afternoon, all built around one goal. Give executives the clarity and the frameworks to lead the AI conversation in their own organisations, starting Monday morning. The day was led by Johnny Than, CEO and Co-Founder of both Appficiency and AskCipher, with over 30 years of enterprise IT experience and more than 1,000 client engagements.
This is a recap of what happened, what was said, and what the room took home.
Enterprise AI Is a Cultural Shift, Not Just a Technology Problem
Johnny opened the morning by drawing a line most executives hadn’t considered: the organisations that survive major technological shifts are rarely the ones that resist them. They are the ones who adapt to how culture changes around the technology.
He walked through the room through a series of historical parallels. Scribes to the printing press. Encyclopedias to Wikipedia. CDs to Spotify. Taxis to Uber. Each disruption followed the same pattern: the technology arrived, created fear and resistance, and then quietly became the new normal.
The question for enterprise leaders isn’t whether AI represents that kind of shift. It does. The question is how to lead through it intentionally rather than reactively.
AI-native organisations are already moving faster because they aren’t locked into rigid approval processes and legacy software workflows. For enterprises, the goal isn’t to chase speed. It’s about redesigning work around the availability of intelligence everywhere, not just automating what already exists.
Why Enterprise AI Keeps Stalling: The Three Faces of GenAI Friction
To make the challenge concrete, Johnny walked the room through one of the most universal enterprise workflows: hiring a great employee.
The exercise covered the full hiring process, from job description creation to resume screening, interview scheduling, interview evaluation, and final candidate selection. At each stage, a different kind of friction surfaced.
Technical Friction
A generic LLM can draft a job description, but it lacks organisational context. It doesn’t know your competency frameworks, your compensation bands, or what “senior operations” actually means at your company. The output may sound polished, but it doesn’t align with how your organisation actually hires, which creates a second review process that slows everything down.
Human Friction
In-app agents can screen hundreds of resumes automatically, ranking candidates by experience, certifications, and keywords. The problem: recruiters don’t trust the output. Hiring managers override recommendations based on instinct. Employees worry about bias. The AI does the work, and humans redo it anyway.
Organisational Friction
AI scheduling tools will book your interview in the first available calendar slot. Nobody wants that. Add multiple panellists, competing priorities, and an urgent posting, and the AI creates more coordination overhead than it removes.
By the time you reach final candidate selection, all three friction types have compounded. The organisation is using AI everywhere in the process and trusting it almost nowhere.
This, Johnny argued, is the real reason most organisations aren’t seeing return on their AI investments. The technology is working. The friction is the problem.
Bite Small Chew Fast: The Enterprise AI Framework That Actually Works
The framework that anchored the entire day came from a simple internal mantra we’ve adopted: bite small, chew fast.
Most organisations read the headlines and decide their first AI initiative should be transformational. They want to solve a £50 million problem with AI. The consistent pattern across enterprise AI deployments is that this approach fails.
The better entry point is to solve a five-hour problem with twenty minutes of AI. That’s how you build confidence, demonstrate value, and create the organisational buy-in that makes larger AI investments possible.
This isn’t a ceiling. There are million-dollar problems that have been solved with a fifty-dollar AI subscription. But attempting that first, before the organisation has developed AI literacy, governance, or trust, leads to exactly the friction described above.
The principle maps to Appficiency’s Solution Levelling Framework, where complexity is measured across five levels, from basic inspection and monitoring at Levels 1 and 2, through targeted automation at Level 3, and into full production-grade automation at Levels 4 and 5. Levels 1 through 3 resolve the majority of business needs. The effort required to reach Level 5 is roughly 25 times greater than Level 1.
Start simple. Prove it works. Escalate complexity only when data justifies it. That’s what bite small chew fast means in practice.
What Real Enterprise AI Success Looks Like
The panel brought three distinct perspectives on what bite small, chew fast looks like in production.
Alex Miles, Partner at 180 Systems, described how AI helped a client achieve an ERP go-live in under two months, a timeline previously considered impossible, by using AI to automate the data migration mapping process and generate report code from existing templates. Two contained interventions. One compressed timeline.
“New companies we’re working with want to get to value much faster,” Alex noted. “If you start with a single process and gain some confidence, the team can trust the system and build on that.”
Shayan Rastgou, Co-Founder of AskCipher, shared that Appficiency’s 200 to 300 daily AskCipher users fall into three use case categories: consultants using it to accelerate ERP implementations, developers using it for code guidance, and employees using it for communication and external-facing tasks. The third category is consistently the most complex, because it requires the kind of institutional context and judgment that AI models don’t naturally carry.
“If we take the time to provide AI with the right context and help humans understand how to interact with it, the implementation becomes a lot less friction-heavy,” Shayan said.
Gareth Doherty, PhD, from Think/able Solutions, offered a more grounded perspective on where most organisations actually are: the organic co-piloting stage. Everyone has access to a large language model. Most are figuring out how to use it. Very few are driving disciplined, measurable enterprise AI value from it yet.
“Adoption means learning as you grow,” Gareth said. “If you’re going to start small, start with your strongest teams. They know their processes better and will give you the most useful signal.”
The Conversations That Mattered: Key Q&A Insights
The Q&A after the panel was where the session found its sharpest edges. The room pushed back, shared real scenarios, and surfaced some of the most practically useful insights of the day.
Reliability Is the Hardest Thing to Prove in Enterprise AI
When asked what is most difficult to establish in an enterprise AI implementation, the panel aligned around one answer: reliability. With a trained employee, once certified, you can trust the output. AI is probabilistic. Even well-defined deployments carry inherent risk, and in enterprise contexts where invoices, compliance records, and financial transactions are involved, that risk is not theoretical.
Building reliability in enterprise AI requires starting small, measuring rigorously, and extending trust incrementally, not deploying broadly and hoping for the best.
Change Management Is Non-Negotiable
An audience member described a company that deployed AI-assisted sales coaching tools and triggered a team revolt. After two months of resistance, the company explained the purpose clearly: not surveillance, but coaching. Not replacement, but enablement. Performance improved significantly once the team understood the intent.
The panel treated this as a textbook case. Communication and transparency are not soft considerations in an enterprise AI deployment. They are structural requirements. Gareth pointed to DBS Bank in Singapore as a leading example, where its AI ethical framework centers on two factors: explainability and respectability.
“Open communication and transparency are key,” Gareth noted. “If people can see and understand how AI adoption will impact them, you can reduce resistance significantly.”
Enterprise AI Amplifies What’s Already There
One of the more candid exchanges surfaced a hard truth: AI tends to amplify low performers more than high performers. A high-performing employee gets incrementally better. A low-performing employee gets a tool that scales their output, including their mistakes. In organisations that deploy AI broadly without governance or guardrails, problems don’t continue at the same rate. They accelerate.
The panel’s consistent answer: start with your strongest teams. Pilot with engaged people who know their processes. Learn from that foundation before expanding.
Making the ROI Case for Enterprise AI Without Perfect Data
An audience member asked directly how to justify AI investment internally when the ROI data is still limited. The honest answer from the panel was that this is a real constraint in 2026, not a perception problem. The data infrastructure to measure enterprise AI ROI at scale doesn’t yet exist in most organisations.
The practical path forward is to anchor your AI initiative to a business process that already has performance metrics. If you can measure the baseline, you can measure the improvement. Government grant programs, including the Ontario Centre of Innovation and ADC’s Lift program, are also available to support organisations at the evaluation stage.
Build vs. Buy in Enterprise AI: Context Wins
The question of whether to build AI tooling internally or purchase it came up toward the end of the session. The panel reframed it entirely: the differentiator in enterprise AI isn’t whether you can build the capability. Anyone can build an LLM-based system that calls tools and holds a conversation. The differentiator is the knowledge and context encoded into that system over time, the field mappings, the workflow logic, the organisational rules, and the edge cases discovered through hundreds of real deployments. That takes years to accumulate.
The Most Important Skill for Enterprise AI Right Now
When asked what individuals and organisations should be developing, the panel pointed to something most people didn’t expect: communication.
“The biggest element of getting what we want out of AI is how well we communicate with it,” Shayan said. “The same problem, communicated differently, produces a completely different result.”
As general knowledge becomes increasingly commoditised through AI models, the ability to communicate precisely and translate organisational context into effective inputs becomes a meaningful competitive advantage. The best prompts don’t come from the most technical people in the room. They come from the people who best understand the work.
A Look at the Afternoon: Your Best Enterprise AI Strategy Has No UI
The afternoon session took a different angle, focused on where enterprise AI is heading rather than where it is today.
Johnny argued that the traditional enterprise interface, the menus, screens, dashboards, and click paths that define how most business software operates, is becoming the primary barrier to enterprise AI adoption rather than its enabler. The most powerful AI strategies don’t add a better interface. They remove the need for one. Context replaces navigation. One sentence replaces dozens of clicks.
AskCipher was demonstrated live as an example: a single AI layer that sits above an organisation’s entire connected application stack, executes multi-step workflows across systems, enforces organisational rules and permissions, and operates without requiring the user to know where anything lives in the underlying software.
Five Enterprise AI Principles to Take Back to Your Organisation
The day closed with five principles presented not as aspirational goals but as architectural requirements for enterprise AI done right.
- Your data stays yours. AI should never own or store your data. Every action happens within your existing stack.
- It’s a layer, not a feature. Enterprise AI isn’t an add-on to one application. It’s a universal interface across your entire stack.
- It enhances, it doesn’t replace. AI works within the capabilities of your existing systems. The systems stay. The experience changes entirely.
- Permissions are sacred. AI should only do what the authenticated user is permitted to do. No escalation. No workarounds. Full audit trail.
- One perfect action beats ten messy ones. Precision matters more than speed. Confirm before acting. Prioritise quality of outcome over volume of output.
Frequently Asked Questions About Enterprise AI
What does “Bite Small Chew Fast” mean in enterprise AI strategy?
Bite Small Chew Fast is the principle that enterprise AI adoption should start with contained, high-impact use cases rather than large transformational initiatives. The goal is to solve a five-hour problem with twenty minutes of AI, demonstrate real value, build organisational confidence, and scale from there. Attempting to solve complex, enterprise-wide problems before the foundation is in place is one of the most consistent patterns behind enterprise AI deployments that stall.
What is GenAI friction, and why does it matter for enterprise AI ROI?
GenAI friction is the resistance that slows or stops enterprise AI adoption. It comes in three forms: technical friction, where AI tools lack organisational context and produce outputs that don’t align with real operations; human friction, where employees don’t trust AI outputs and revert to manual processes; and organisational friction, where gaps in governance, policy, and cross-departmental coordination prevent AI from scaling. Identifying which type is most present in your organisation is the first step toward addressing it and unlocking real AI ROI.
How is a context-driven AI layer different from tools like ChatGPT or in-app enterprise AI agents?
Generic AI tools operate on public knowledge and have no memory of your organisation, your workflows, or your previous interactions. In-app agents are limited to a single application and typically don’t enforce organisational permissions. A context-driven AI layer operates above your entire connected application stack, executes multi-step workflows across systems, retains organisational context across sessions, and enforces each user’s existing permissions. The difference isn’t capability. It’s context.
How do you make an ROI case for enterprise AI when the data is still limited?
The most practical approach is to anchor your AI initiative to a business process that already has performance metrics. If you can measure baseline performance, you can measure AI’s impact. Starting with one contained process also reduces the governance and change management complexity that makes broader enterprise AI deployments harder to justify. Government programs, including the Ontario Centre of Innovation and ADC’s Lift program, are also available to support organisations at the evaluation stage.
The Advantage Won’t Come from Moving Faster
The consistent thread across every session, every panel exchange, and every audience question was the same.
Speed is not the competitive advantage in enterprise AI. Intentionality is. One well-designed AI deployment, built on clean data, governed permissions, and a clear process foundation, creates more durable value than ten rushed experiments.
As Johnny put it to close the day: “The advantage won’t come from adopting AI faster. It will come from applying it more intentionally than everyone else.”
Bite small. Chew fast. Build something that actually lasts.
To explore how Appficiency, AskCipher, NimbusPoint and SnapOn Software can help your organisation move from AI experimentation to real operational impact, book a strategy call with our team today.