CIRA IQ V2: From Automation to an AI Operating Layer

From Automation to an AI Operating Layer

We built the first version to help businesses automate their work. Then our users showed us that automation was only the beginning.

There is a moment in every technology company when the product stops being entirely yours. It happens when the roadmap meets reality. You can spend months defining the problem, designing the interface, writing the architecture and deciding what the product should become. Then you put it in the hands of real people.

People do not use the product exactly the way you imagined. They build things you expected. They build things you never considered. They find the edges of the system. They ask questions the original product was not designed to answer.

If you are paying attention, those questions are not distractions. They are product strategy.

That is what happened with CIRA.

We started by building automation. The idea was straightforward. Businesses already have processes. Those processes repeat. Information moves from one place to another. People copy, paste, check, update, notify, follow up and report.

Why should a person have to keep doing the same work?

So we built CIRA around workflows. Take a process. Turn it into a workflow. Connect the tools. Let the system execute it.

It worked.

Then people started asking for something more fundamental. Not “Can CIRA automate this?” but “Can CIRA understand what I am actually trying to do?”

That question changed the direction of the product. It is the reason CIRA V2 exists.

The software problem nobody solved

For years, businesses have responded to complexity by buying more software. There is software for email, sales, accounting, customer support, communication, documents, analytics and now artificial intelligence.

We have never had more digital tools. Yet the person running the business is still often the system connecting them.

A customer sends an enquiry. Someone reads it. Someone decides what it means. Someone enters it into a CRM. Someone assigns it to a salesperson. Someone remembers to follow up. Someone checks whether the salesperson followed up. Someone eventually puts the numbers into a report.

The applications are digital. The operation is still human.

This is the paradox of modern business software. We have automated individual pieces of work without necessarily making the business itself intelligent.

CIRA’s first version addressed part of that problem. But our users showed us something important. They did not want another place to configure technology. They wanted technology that could understand the work.

Businesses do not think in nodes

This sounds obvious, but it has enormous implications. A business owner does not wake up thinking about nodes. They do not think “Connect Gmail to an AI classifier, map the output to a database, trigger a notification and create a follow-up task.”

They think “Every new customer enquiry should be understood, recorded and followed up.”

That is the difference between a workflow and an intention. Traditional software asks people to understand the system before they can tell it what they want. AI gives us the opportunity to reverse that relationship. The user can start with the outcome. The system can increasingly figure out the work required to get there.

That is one of the biggest ideas behind CIRA V2. The workflow has not disappeared. It has become the foundation beneath a more intelligent system.

When the workflow starts thinking

This is where the distinction between automation and intelligent automation becomes important. A traditional workflow follows instructions. If this happens, do that. Then do this. Then send that. It is predictable. It is useful. And for many processes, it is exactly what you want.

But businesses do not operate entirely through predictable instructions. Information changes. Customers behave differently. Messages arrive in unexpected formats. Decisions depend on context. Sometimes the next step is not obvious until the system understands what just happened.

That is where AI becomes useful. Instead of AI living in a separate tab, CIRA V2 brings intelligence into the workflow itself. A message can be interpreted. Information can be classified. A decision can be informed by context. The next action can be determined. And the workflow can continue.

The difference is subtle but important. Automation follows the process. Intelligence understands the process. When the two come together, software starts becoming capable of participating in the work rather than simply executing instructions.

The canvas is becoming a way of thinking

We also learned that building automation should not feel like programming a machine. That is part of why we built the Smart Canvas. The canvas gives businesses a visual space to bring their tools, information and intelligence together. You can put an application on the canvas, bring in AI, connect the pieces and see how information moves. You build around the actual operation.

The deeper idea is not the visual interface. It is that software should increasingly reflect how people think about work. A business is not a collection of disconnected applications. It is a collection of activities, decisions, information and outcomes. The interface should help people see those relationships.

The canvas is our step in that direction.

Then AI agents changed the question

The conversation around AI has moved quickly. A few years ago, the question was whether AI could generate an answer. Then whether AI could reason about a task. Now the question is whether AI can actually do the work.

That is where agents become interesting. It is also where things become dangerous. Giving an AI system the ability to act is fundamentally different from giving it the ability to generate text. A wrong paragraph can be edited. A wrong email can be sent. A wrong database update can propagate. A wrong financial transaction can cost real money.

An agent that can act without understanding the consequences is not intelligent enough simply because it is autonomous.

So when we built agents into CIRA, we were not interested in autonomy as a marketing word. We were interested in controlled execution.

Autonomy without guardrails is a liability

CIRA’s agents can work with the capabilities already available across the platform. They do not need users to manually wire every possible tool into the agent. The agent can reason through a goal, select a tool, inspect the result, decide what should happen next and replan when necessary.

But capability is not permission. Before an action happens, CIRA evaluates it. Is the action read-only? Can it be reversed? Or is it irreversible?

That distinction matters. Reading a spreadsheet is not the same as sending an email. Updating a record is not the same as deleting it. Sending money is certainly not the same as checking a balance.

CIRA’s decision layer considers both confidence and risk. The result can be to execute, ask for approval or decline. We deliberately made one principle difficult to bypass. Irreversible actions require a human.

Not because we do not believe AI will become capable enough. Precisely because we believe it will become capable enough to do things that matter.

When approval is required, the agent does not simply stop. The workflow can pause. An approval request can arrive in Telegram. Approve it and the execution continues from where it stopped. Reject it and the action does not happen. The decision becomes part of the audit trail.

There are also boundaries around what an agent can do. Allow and deny lists, rate and spend limits, and kill switches that can stop an agent or all agents owned by an organisation.

This is the part of agentic AI that does not always make the demo. The exciting part is the agent deciding what to do. The difficult part is building the machinery that decides what the agent is allowed to do.

That is where real-world AI begins.

Memory is powerful. Context is more important.

The moment an AI system starts doing work repeatedly, another problem appears. Memory. A useful business agent cannot behave as though every interaction is the first interaction. It needs context.

But giving an AI system unlimited memory is not necessarily intelligence. It can be a security problem. It can be a privacy problem. It can simply be confusing.

So CIRA treats memory as something with boundaries. Memory can belong to a workflow. It can belong to a user. It can belong to an organisation. Or an agent can remain stateless.

That distinction matters because context is not just information. Context is permission. An agent working on one workflow should not automatically inherit everything another agent has learned. A team agent may need organisational knowledge. An individual agent may need personal preferences. A workflow may only need to remember what happened within that process.

The future of business AI will depend not only on what systems remember, but on whether they understand what they are supposed to remember.

Voice should not create a different intelligence

There is another lesson here that seems small until you think about it. People do not always work at a keyboard. They send voice notes. They speak while travelling. They dictate instructions between meetings.

In CIRA, voice is not treated as a fundamentally different type of agent. A voice instruction becomes text at the edge. From there, it enters the same system. Same agent. Same memory. Same tools. Same permissions. Same approval rules.

The interface changes. The underlying intelligence does not.

That matters because the future of business software probably will not be defined by one interface. It will be defined by whether the underlying system can understand intent wherever that intent comes from.

The workflow becomes a product

Perhaps the most interesting shift in V2 is what happens after someone builds something useful. Historically, automation has been personal. You build a workflow for your company. You solve your problem. Then the value largely stays inside your organisation.

But what if the thing you built could become useful to someone else?

That is the thinking behind the CIRA Marketplace. A workflow can be published. Someone else can discover it. They can get their own copy, configure it with their own credentials and adapt it to their own business. The creator can earn from what they have built.

This changes the role of the workflow. It is no longer only an internal automation. It can become a reusable piece of operational infrastructure.

This creates an interesting progression. Workflow becomes intelligent workflow. Intelligent workflow becomes agent. Agent becomes product.

AI agents do not have to be things that companies merely consume. They can become things people build. And eventually, things people sell.

That is a much bigger idea than an automation template library. It starts to look like an economy around operational intelligence.

Intelligence needs to know what happened

There is another piece of the puzzle that becomes important as automation becomes more autonomous. Measurement.

If software is increasingly doing work on your behalf, you need to know how that work is performing. Not just whether the software is technically running. Whether the operation is actually working.

CIRA V2 introduces weekly analytics built around immutable snapshots of workflow performance. Runs. Success rates. Duration. Slowest steps. The system creates a permanent record of the week rather than continually rewriting history.

That may sound like a small engineering decision. It is not.

If an organisation is going to trust software with increasingly important work, it needs to be able to look back and ask what happened, how it performed, what changed, where time is being lost and where the operation is breaking down.

Operational intelligence begins with operational truth.

The pieces are beginning to connect

If you look at each V2 capability independently, it is easy to see a collection of features. Smart Canvas. AI models. Agentic Mode. Memory. Integrations. Approvals. Analytics. Webhooks. Marketplace.

We do not think about them as separate features. They are pieces of a larger system.

The progression we are building toward is business intent to workflow to data to intelligence to decision to action. The workflow provides structure. Operational data provides context. AI provides intelligence. Memory provides continuity. Agents provide execution. Guardrails provide control. Analytics provide visibility. The Marketplace allows what is built to travel.

Put together, these capabilities begin to look less like an automation product and more like an operating layer for business.

That is the direction we are heading.

We did not build V2 because we wanted more features

This distinction matters to us. V2 was not born from a desire to release a longer changelog. It came from using the first version in the real world.

Our users showed us what worked. They showed us where building was difficult. They showed us that workflows were only useful if they reflected the actual way businesses operate. They showed us that AI needed to move closer to the work. They showed us that agents needed controls, not just capabilities. They showed us that what one person builds could become valuable to another.

So V2 is not a rejection of what came before. It is the next layer. The workflow is still the foundation. We are simply making that foundation more intelligent.

The next question is not what AI can tell us

This is where the conversation gets interesting. Imagine asking your business system why sales dropped this month. A traditional dashboard gives you a number. A chatbot gives you an explanation. An intelligent operating layer could go further.

It could examine the operational data. Understand the workflows involved. Identify a pattern. Connect that pattern to what has changed. Recommend an action. And, where authorised, take that action.

That is a fundamentally different relationship with software. The system is not simply answering a question. It understands enough of the operation to participate in what happens next.

That is the future we are building toward. Not AI as another application. Not automation as another collection of integrations. But intelligence embedded into the way a business actually works.

From automation to an AI operating layer

CIRA started with a simple question. What if businesses did not have to keep doing repetitive work manually?

V2 asks a bigger one. What if the system doing that work could actually understand it?

That is the journey from automation to intelligence. From intelligence to decisions. From decisions to action. And eventually, from individual workflows to an AI operating layer that understands how a business operates and helps run it.

We are still early. There is much more to build for businesses. But the direction is clear.

The future of business software may not be another application you open every morning. It may be a layer that sits across the operation, understanding intent, working with your tools, learning context, watching what is happening, making decisions and taking action within the boundaries you set.

That is what we are building with CIRA.

CIRA V2 is live. You do not need to automate your entire business. Start with one process. One repetitive task. One thing your team keeps doing that software should probably be handling.

Build it. Connect it. Make it intelligent. See what happens when the software stops being somewhere you go to do work and starts becoming part of the work itself.

Go build something.

https://ciraiq.com