The Harness Matters More Than the Horse

What Does a Fox Hunt Have to Do with Artificial Intelligence?

Some memories stay with you because they are so unusual that your brain seems to file them away in high definition.

Years ago, a friend of mine was on the marketing team for a fox hunt and invited me along.  And no, I was not riding a horse.

I was standing with the crowd near the beginning of the event, surrounded by riders, horses, spectators, and all the pageantry you would expect from something that felt transported from another century.

Then there was the champagne.

Someone had constructed this enormous display of champagne glasses, stacked one on top of another in a shimmering tower. It was one of those elaborate touches designed to make the whole occasion feel even grander.

The riders gathered.  The horses shifted beneath them.  The priest began to bless the hunt.  And then everything went wrong.

One of the horses became unwieldy.  The rider tried to regain control but couldn't.  In a matter of seconds, this magnificent animal crashed into the tower of champagne glasses. Glass went everywhere. The rider came down with the horse and was injured.

The elegant ceremony suddenly became chaos.

I remember thinking afterward about the extraordinary amount of power contained in a horse.  Under control, that power is graceful.  Directed properly, it can accomplish something extraordinary.  But power without sufficient control can become dangerous remarkably quickly.

Today, as artificial intelligence races into the enterprise, I keep thinking about that horse.  Because we may be spending far too much time talking about the horse and not nearly enough time talking about the harness.

What is an AI Harness?

An AI harness is the surrounding system that allows an AI model to operate safely, effectively, and productively in the real world.

For the first few years of the generative AI revolution, nearly everything revolved around the models.  Which model was smartest?  Which had the largest context window?  Which reasoned better?  Which produced the best code?  Which was fastest?  Which scored highest on the newest benchmark?

Those questions still matter.

But increasingly, AI engineers and commentators are talking about something else: the harness around the model.

The model provides intelligence.  The harness determines how that intelligence is used.

An AI harness can provide context, connect the model to tools and data, establish permissions, define workflows, validate outcomes, create feedback loops, preserve memory, and determine when humans should become involved.  

That distinction becomes critical inside a large enterprise.  For an individual using an AI assistant, some of these capabilities may simply be conveniences.  For a bank, airline, hospital, telecommunications provider, government agency, or other large enterprise, they are necessities.

Why Do Large Enterprises Need a Different Kind of AI Harness?

Because the consequences, complexity, and scale of AI are dramatically greater inside a large enterprise.

Imagine an AI assistant helping someone decide what to cook for dinner.  Perhaps something is wrong.  Maybe it recommends a restaurant that is closed.  Maybe it forgets that you don't like Thai food.  Maybe it hallucinates an ingredient.

Annoying?  Absolutely.

Catastrophic?  Probably not.

Now place AI inside one of the world's largest banks.  Or airlines.  Or healthcare organizations.  Or governments.  Or telecommunications providers.

Suddenly the same intelligence may have access to customer identities, account information, payment systems, medical information, employee systems, regulatory processes, communications infrastructure, and millions of customer interactions.

The horse just became considerably larger.  And considerably faster.

An AI system capable of taking action across an enterprise could potentially change an account, issue a refund, schedule an appointment, alter a reservation, retrieve private information, trigger a workflow, contact a customer, authenticate a user, update a database, or make thousands of decisions simultaneously.

That is precisely why enterprise AI cannot simply be a bigger version of consumer AI.  Large enterprises need a different kind of harness.  One built for scale.  One built for complexity.  One built for security and governance.  One built for resilience.  One built to preserve context.  And perhaps most importantly, one built to work with the extraordinarily complicated technology environments enterprises already have.

What Makes an Enterprise AI Harness Different?

A true enterprise AI harness must do far more than connect an AI model to an application. It must connect intelligence to the enterprise's entire operating environment.

That includes:

• customer context
• enterprise data
• communications
• business applications
• security policies
• permissions
• workflows
• compliance requirements
• human employees
• multiple AI models
• legacy systems
• cloud environments
• on-premises infrastructure
• business continuity and resilience

This is where the conversation becomes particularly interesting for CX orchestration.

What Role Does Avaya Play in the AI Harness Conversation?

Avaya has spent decades operating inside some of the largest and most complex enterprise environments in the world.

There is an irony here.  The technology industry is suddenly discovering how important orchestration, context, reliability, governance, and integration become when intelligent systems interact with the real world.  Avaya's largest customers have been dealing with variations of those challenges for decades.

Think about what happens when someone contacts a major airline because a flight has been canceled.  That interaction might involve the contact center, loyalty systems, reservations, payment platforms, flight operations, identity systems, mobile applications, airport personnel, partner airlines, customer history, and increasingly AI.

Or consider a customer contacting a major bank because of a suspicious transaction.  Understanding the customer's sentence is the easy part.  The system may need to establish who the customer is, understand the customer's history, retrieve the appropriate accounts, identify potential fraud, apply regulatory rules, authenticate the customer, recommend an action, invoke another enterprise system, document what happened, and determine whether a human should take control.

That isn't simply a chatbot problem.  It is an orchestration problem.  It is a context problem.  It is a trust problem.

It is a harness problem.

And that is exactly one of the critical roles that Avaya Infinity is designed to play.

Should Avaya Infinity Be Thought of as Providing an Enterprise AI Harness?

Yes. One useful way to understand Avaya Infinity is as an enterprise CX harness connecting AI with systems, data, workflows, customer context, and people.

The strategic value of Avaya Infinity is not simply that it incorporates AI.  Its value is that it can help provide the environment in which enterprise AI becomes useful.

Avaya Infinity is designed to help large organizations connect customer interactions with the systems and intelligence surrounding them.

That changes the discussion.  The question is no longer simply:

How smart is the AI?

The more important question becomes:

Can that intelligence safely accomplish something useful inside the enterprise?

Why Does Customer Context Matter So Much to Enterprise AI?

Because even the world's smartest general-purpose AI model does not automatically understand the history, relationship, permissions, and circumstances surrounding an individual customer.

It does not inherently know that the person calling has been a customer for 17 years.  It does not know they attempted to resolve the problem through the mobile app ten minutes ago.  It does not know that they spoke with another employee yesterday.  It does not know that their shipment has already been delayed twice.  It does not know what their contract permits.  It does not know what your organization promised them.

Unless something gives the AI that context, customer journey context becomes part of the AI harness.  AI provides intelligence.  The enterprise CX environment provides situational awareness.  And the difference between those two things can be enormous.

Why Does an Enterprise AI Harness Need Governance?

Because understanding what should happen and having permission to make it happen are two entirely different things.

Suppose AI determines that refunding $5,000 to a customer would be the most appropriate solution.  Should it?  That depends.

Does the employee have authority to approve it?  Does the AI?  Does the transaction require additional authentication?  Does it trigger a regulatory process?  Is there evidence of fraud?  Does another system have to authorize the transaction?  Should a supervisor become involved?

A sophisticated AI model may understand the customer's problem perfectly.  But understanding something and having the authority to act upon it are not the same thing.  This is why enterprise AI needs orchestration, permissions, governance, auditability, security, and human checkpoints.

The harness doesn't just help AI do things.

The harness determines what AI should and should not be allowed to do.

That may ultimately prove to be one of its most important functions.

Why Does Resilience Matter in Enterprise AI?

Because large enterprises cannot assume that every system, network, integration, or AI model will work perfectly all the time.

There was another thing that struck me about the horse at the fox hunt.  Everything seemed perfectly controlled until it wasn't.

Enterprise technology lives in that world.  Systems fail.  Networks fail.  Models fail.  Cloud services fail.  Integrations fail.  Traffic suddenly spikes.

A perfectly ordinary Tuesday can become a crisis in seconds.   For a small organization, an interruption may be frustrating.  For a major airline, hospital, bank, government agency, or emergency operation, communications can be mission-critical.

That makes resilience part of the AI conversation.

An enterprise AI harness cannot be designed only for the moment when everything works perfectly.  It must account for the moments when it doesn't.

This is particularly relevant to Avaya because our heritage has always included large-scale, mission-critical enterprise communications environments where reliability and continuity matter.

Avaya Infinity extends this heritage into an AI-driven world.

Why Should an Enterprise AI Harness Be AI-Agnostic?

Because no enterprise can know which AI model will be best two, three, or five years from now.

Imagine building your entire enterprise AI architecture around today's best model.  Which would you choose?

Now ask the harder question:

Which model will you choose in 2028?

Nobody knows.  The pace of AI development makes that unknowable.  A model that is extraordinary today may be surpassed by one that is faster, cheaper, safer, more specialized, more private, or dramatically more capable.  Think about this – just a few months ago no one had heard the model name Fable.

Different models may also prove better for different jobs.  One might be best for reasoning.  Another for voice.  Another for translation.  Another for summarization.  Another for healthcare.  Another for fraud detection.  Another for autonomous action.

The enterprise AI harness therefore cannot be permanently welded to the horse.

It needs to accommodate different horses.

This is why Avaya Infinity's emphasis on an open and AI-agnostic architecture is of critical importance to our customers.  The goal is not to force the enterprise to make a permanent bet on one AI provider.  The goal is to allow organizations to select the intelligence appropriate for the use case while preserving the enterprise environment around it.

What Does “CX Modernization” Mean in the Real World?

It means enterprises should be able to introduce new AI capabilities without ripping out the technology, workflows, and infrastructure that already run their businesses.

Large enterprises are not blank sheets of paper.  They have decades of technology investments.  They have applications.  They have workflows.  They have data.  They have compliance requirements.  They have communications environments.  They have thousands or millions of customers.  They have processes that cannot simply be turned off while someone constructs a shiny new AI architecture.

When it comes to large enterprise tech modernization, including adding AI capabilities, the objective is not to throw away everything that came before.  It is to harness new intelligence while preserving what still matters.  That is fundamentally different from the rip-and-replace philosophy that some CX software vendors may be promoting.

What Role Could MCP Play in the Enterprise AI Harness?

Model Context Protocol, or MCP, can help provide a standardized connection between AI systems and enterprise tools, data, APIs, and context.

Think of it as part of the hitch connecting the horse to the carriage.  Without standardized connections, enterprises can find themselves building custom integrations between every model and every system.  That quickly becomes enormously complicated.

MCP offers a more standardized way for AI systems to discover and interact with enterprise capabilities.  Avaya's support for MCP therefore fits naturally into the harness concept.

Instead of asking:

"Which AI does Avaya use?"

Avaya’s customers can ask:

"Which AI would you like Avaya Infinity to orchestrate?"

This is a fundamentally different architectural philosophy.

What Is the Difference Between AI Automation and AI Orchestration?

Automation follows predetermined rules. Orchestration coordinates multiple systems, data sources, AI capabilities, workflows, and people around an outcome.

For years, enterprises have automated things.  Press 1.  Fill out this form.  Follow this workflow.  If X happens, do Y.

Generative and agentic AI change the equation because the system can increasingly interpret situations rather than simply follow predetermined rules.  But that makes orchestration more important, not less.

An intelligent system must understand not merely what happened but what should happen next.  Which system should participate?  What information should travel with the interaction?  Which AI model should be invoked?  Which business rule applies?  When should another workflow begin?  When should a person take over?  

That is why enterprise AI requires more than automation.  It requires orchestration.

And it is leading Avaya’s large installed base of CX software customers to think about Avaya Infinity as…   

… the harness between enterprise intelligence and enterprise action.

Customer on one side.  Enterprise on the other hand.  Between them sits an orchestration layer that connects conversations, customer context, enterprise data, workflows, AI models, applications, and people.

Not AI replacing the enterprise.

AI harnessed by the enterprise.

Will the Company with the Most Powerful AI Model Win?

Probably not.  At least, not automatically.

This brings us back to that fox hunt.  What I remember most isn't the champagne glasses.  It isn't even the horse.  It is the speed with which something powerful and elegant becomes something dangerous.  One moment, everything was controlled.  A few seconds later, the rider was on the ground.

There is a lesson there for the AI era.  We understandably marvel at increasingly powerful models.  Their ability to reason, speak, see, listen, create, and act is extraordinary.  And they will only become more capable.

But as the horse becomes stronger, the quality of the harness becomes more important, not less, particularly for the world's largest enterprises.

Organizations are responsible for millions of customers, critical infrastructure, enormous amounts of sensitive information, and customer interactions where getting things wrong can have very real consequences.

These organizations don't just need access to AI.  They need AI that can operate within their world.  AI connected to the right data.  AI given the right context.AI governed by the right policies.  AI connected to the right workflows. AI working alongside the right people AI capable of using whichever models are best for the job.  AI deployed without forcing an enterprise to abandon the systems its business already depends on.  AI that remains reliable when the unexpected happens.

They need an enterprise AI harness.

So how should we all think about Avaya's role in the AI era?

For decades, Avaya has operated inside some of the largest and most complicated enterprises in the world, helping connect people, communications, applications, workflows, and customer experiences at enormous scale.

Avaya Infinity extends that role into an AI-first world.  Not by betting the enterprise on a single model.  Not by asking customers to discard what they already have.  But by creating an open enterprise CX platform capable of connecting systems, data, workflows, humans, and whichever forms of AI come next.

The AI industry will continue producing more powerful horses.  That seems inevitable (and even accelerated by new practices such as recursive self-improvement.)  

The more interesting question for the world's largest enterprises may be:

Who is building the harness?

Take a look at this brief overview of Avaya’s unique approach to harnessing the power of AI for the largest enterprises in the world.