Broadcasting from Melbourne: From AI Assistance to Autonomous Execution

Charles Skamser

Welcome back. I am coming to you from Melbourne, Australia this week. I left Boise Saturday night and eventually found myself trying to fit my six foot six frame into a United Polaris pod for almost sixteen hours between San Francisco and Melbourne. And I actually want to give United some credit here. At six foot six, even a Polaris pod is a tight fit when I try to completely stretch out, but compared with trying to make that trip in a normal airline seat, it is fantastic. There is something strange that happens when you have been on the same airplane for that long. You eat, you work, you watch something, you sleep, you wake up, you look at the flight map and wonder why Australia still seems impossibly far away. You work some more, you sleep again, somebody brings you another meal, and eventually you are not completely sure whether it is breakfast, lunch, or dinner. Then you cross the International Date Line and Monday more or less disappears from your lived experience. But while I was somewhere over the Pacific trying to determine what day it was, the AI industry certainly was not standing still.

Catherine Spencer

It, it, it really is a fascinating psychological experiment, Charles. You step off the grid for sixteen hours, suspended in that quiet airplane bubble, and when you reconnect on the other side, the entire enterprise conversation has moved another inch forward. And in fact, while you were over the Pacific, OpenAI released their August 2026 report highlighting what they call the shift from assistance to execution. That transition is really the heart of everything we are seeing in London and across Europe right now.

Edward Hamilton

Yes, absolutely. The assistance to execution pivot is extraordinary. For three years, we have been obsessed with generative tools that help an individual write an email, summarize a contract, or draft some code. But an agentic system is entirely different. It does not just suggest text. It takes direct, autonomous action inside enterprise systems like SAP, Salesforce, and ServiceNow.

Charles Skamser

And, and that is where the airplane metaphor gets really interesting to me. When you look at how individuals use AI assistants, it is very bursty. You log in, you ask a prompt, you get an answer, you log off. But when you move to agentic execution, these workloads run in continuous loops. While I was sleeping over the Pacific, autonomous agents across dozens of enterprises were running overnight batch processes, reconciling ledger items, routing supply chain orders, and invoking APIs without a single human sitting at a keyboard. It is literally always-on digital labor.

Catherine Spencer

Wait, Charles, so when you finally landed in Melbourne and had some time to reflect on what you had read or watched on the flight, was there a specific piece of news that made you realize just how fast this execution phase is taking hold?

Charles Skamser

Well, it was really seeing the telemetry data coming out of these enterprise deployments. OpenAI reported that their most intensive enterprise users are not just generating more text. They are running persistent, multi agent orchestrations that connect organizational context directly to tools and repeatable business workflows. The sheer volume of automated actions taking place behind the scenes while people sleep is staggering. It makes you realize that enterprise AI is finally growing up.

Edward Hamilton

Indeed. And what strikes me, Charles, is the profound contrast between airplane downtime, where time feels completely suspended, and the relentless, non stop iteration of these agentic loops. You lose Monday entirely because of the International Date Line, but digital labor does not recognize time zones or calendars at all. It just keeps executing.

Charles Skamser

Exactly, Edward. And that sets up the big set of questions we are tackling today. Because once you move from asking what can generative AI write to asking what can autonomous agents actually execute, the hard questions hit you like a brick wall. What does all of this cost? Can I govern it? Can I trust it? Who or what is authorized to take an action inside my company? And ultimately, is my organization actually ready to operate this way? Those are the exact questions that have been consuming our team at PX42 for the last year.

Catherine Spencer

And those are precisely the questions that many executives are still completely unprepared to answer.

Catherine Spencer

Let us unpack this assistance versus execution distinction a bit deeper, because I think there is a massive adoption fallacy playing out in the market right now. If Company A buys fifty thousand Microsoft Copilot licenses and hands them to employees to draft emails or polish presentation slides, and Company B redesigns twenty core operational processes around autonomous agents that execute transactions, on paper, both companies have adopted AI. But organizationally, structurally, they are on entirely different planets.

Edward Hamilton

Mm, completely. Giving an employee an AI writing assistant is fundamentally like handing them an advanced electronic calculator. It speeds up individual thought, but the human remains the sole operator, the sole filter, and the sole legal actor. But delegating execution to an autonomous agent is like giving a digital worker limited power of attorney over your core enterprise resource planning and customer relationship management systems. The agent can read records, update databases, issue invoices, or alter customer accounts without intermediate human review.

Charles Skamser

I, I, I love that analogy, Edward. Limited power of attorney is spot on. And that brings us directly to what we call the AI readiness paradox. So many enterprise executives think that because they signed a enterprise agreement for Agentforce, or ServiceNow agents, or ChatGPT Enterprise, that they are somehow AI ready. But owning the software license is not readiness. AI readiness is whether your enterprise possesses the data lineage, zero trust security, identity governance, authority structures, and operational observability required to let an autonomous agent act safely on your behalf.

Catherine Spencer

And Charles, that is where we see so much frustration among European and UK enterprise leaders right now. I was speaking with a chief technology officer of a major financial institution in London last week. They bought thousands of licenses for a leading agent platform. They were excited to automate end to end loan processing. But the moment they attempted deployment, everything ground to a halt. Why? Because their underlying data access permissions were so messy that if an agent was granted access to run the workflow, it inadvertently exposed confidential wealth management records across unauthorized departments.

Edward Hamilton

Ah, the classic enterprise permission leakage problem. In a traditional setup, human role based access controls are sloppy, but human social norms prevent people from searching for things they should not see. An autonomous agent, however, will exhaustively query every database it has technical permission to read. If your data governance is brittle, agentic execution becomes an operational hazard.

Charles Skamser

That is precisely why we developed the PX42 AI Agent Readiness Index. We wanted to give enterprise leaders a rigorous, multi dimensional diagnostic tool that goes far beyond asking how many licenses did you buy. We look at ten distinct dimensions, including data architecture, identity and authority governance, runtime observability, security guardrails, workforce skills, and economic visibility. It gives executive teams an objective benchmark of whether they can safely transition from assistance to execution, and how they compare against industry peers.

Catherine Spencer

So Charles, when you walk a client through the AI Agent Readiness Index, what is the single biggest gap that usually shocks C suite executives?

Charles Skamser

It is almost always authority and identity governance. They simply have not thought through what happens when an agent needs to invoke another agent to complete a transaction. Who authorized that secondary agent? What identity does it hold? What happens if it makes a mistake? They realize very quickly that their existing identity management systems, built for human employees logging into web portals, are completely inadequate for managing populations of autonomous software entities.

Edward Hamilton

It is the structural shift from managing human tools to managing a digital workforce. And if you do not have the operational control plane to govern that workforce, you are essentially launching digital workers into your core systems without job descriptions, audit trails, or clear managerial supervision.

Edward Hamilton

Now, if governance and readiness are one side of the enterprise coin, the other side is economics. And frankly, the recent numbers coming out of the industry are astonishing. The KPMG Q2 2026 AI Pulse Survey surveyed over two hundred C suite leaders at companies with over a billion dollars in revenue. They found that enterprise organizations plan to invest a weighted average of two hundred two million dollars in AI over the next twelve months. Two hundred two million dollars!

Catherine Spencer

And Edward, here is the kicker from that exact same KPMG study: while they are committing over two hundred million dollars on average, only twenty six percent of those organizations report having full, real time visibility into what running AI systems actually costs at scale. Twenty six percent! That means three quarters of major enterprises are deploying massive capital into AI without any real time financial dashboard for operational running costs.

Charles Skamser

Think about that for a second! Imagine going to your Chief Financial Officer and saying, hey, we are going to spend two hundred two million dollars over the next year on a critical technology capability, but once it is running, only one out of four of us will be able to tell you in real time what our operational bills actually look like. How long do you think a CFO is going to tolerate that kind of financial opacity?

Edward Hamilton

Not long at all, Charles! In fact, thirty five percent of leaders in that KPMG survey explicitly identified cost management and economic literacy, specifically understanding usage based pricing like token and inference costs, as a primary barrier to AI adoption. And the root cause is that people are still trying to evaluate agentic AI costs using traditional software cost mental models.

Catherine Spencer

Could you explain that mechanism, Edward? Why does traditional software pricing completely break down when applied to multi agent systems?

Edward Hamilton

Well, with traditional software as a service, you pay a fixed fee per seat, say fifty dollars per user per month. Your financial spend is linear and predictable. In simple generative AI, a user sends a prompt, gets a response, consuming maybe a few thousand tokens. But when you deploy an agentic system, you trigger a multi agent token multiplier. A user asks an agent to solve a complex problem. The primary agent reasons about the task, breaks it into subtasks, queries a database using retrieval augmented generation, calls three external software interfaces, delegates a subtask to a specialized secondary agent, receives an error, retries the step, evaluates the result, validates the formatting, and finally presents an answer.

Charles Skamser

That single initial user request might result in forty or fifty underlying model calls! And if that agent gets stuck in an recursive loop because an external API returned an unexpected error format, it might execute hundreds of retries in a matter of minutes. That is the ultimate CFO nightmare scenario: unthrottled cloud bill shock over a weekend because an autonomous workflow got caught in an infinite retry loop without automated circuit breakers or budget caps.

Catherine Spencer

It is like leaving a tap running in a building you only visit once a month. By the time you receive the water bill, the basement is completely flooded and you owe millions of dollars.

Edward Hamilton

Precisely. Plainly put, agentic AI cannot be budgeted like a fixed software license. It must be budgeted like a variable utility, similar to electricity or raw compute, where consumption fluctuates dynamically based on task complexity, system friction, and workflow retry behavior.

Charles Skamser

And that is why we keep hammering on the fact that AI economics is about to become a mandatory enterprise management discipline. If you do not have granular, real time cost monitoring built into your agent architecture, you are operating with a massive two hundred million dollar blind spot.

Catherine Spencer

To address this economic challenge, professional services firms like EY have started advocating for a broader framework they call the Total Cost of Agents. And at PX42, Charles, you have taken this concept even further with our framework for Digital Labor Cost. Can you walk us through how Digital Labor Cost shifts the entire financial discussion?

Charles Skamser

Gladly, Catherine. For the last three years, everyone has been obsessed with model token pricing. How much does a million tokens cost on GPT 4, or Claude, or Gemini? While token prices matter, focusing exclusively on token cost is like trying to calculate the total cost of an employee by looking only at their hourly wage, while ignoring healthcare, equipment, taxes, management overhead, office space, and administrative support.

Edward Hamilton

Right. Digital Labor Cost looks at the fully loaded cost per unit of completed business work. It asks: what did it actually cost the enterprise to process this specific insurance claim, draft this financial report, or resolve this customer service escalation through digital agents?

Charles Skamser

Exactly. To calculate true Digital Labor Cost, you have to account for the entire operational stack. That includes model inference costs across primary and secondary models, API gateway throughput fees, orchestration platform overhead, vector database storage, human in the loop supervision time, failure recovery costs, and regulatory logging compliance. When you add all those components together, you get the real economic cost of that digital worker.

Catherine Spencer

And once you measure work in terms of Digital Labor Cost, it completely transforms your system architecture, does it not? You realize that using your most massive, expensive frontier model for every single step of a workflow is financial suicide.

Charles Skamser

It really is! Why use a massive seventy billion parameter frontier model to perform basic data extraction or validate JSON formatting when a specialized, lightweight eight billion parameter domain model can do that specific step faster, cheaper, and just as accurately? In our optimization work at PX42, using frameworks like AgentPrune and SafeSieve, we have shown that intelligent model routing, microVM execution runtimes, and dynamic context pruning can reduce total Digital Labor Cost by up to seventy percent without sacrificing output quality or accuracy.

Edward Hamilton

Seventy percent reduction is dramatic! It is the difference between an agent deployment being economically viable or financially ruinous at enterprise scale.

Charles Skamser

I was actually having dinner here in Melbourne with executive leaders from major Australian banking and mining organizations. What fascinated me is how they are starting to evaluate digital agents alongside human full time equivalent headcount in their corporate capital allocation budgets. They are no longer asking how much software do we buy. They are asking: if we deploy digital labor in this business unit, what is its fully loaded cost per transaction compared to human labor, what is its reliability, and what is its net return on investment?

Catherine Spencer

That is a profound shift. It moves AI from being viewed as an IT software expense item to being managed as a core operational workforce capability.

Edward Hamilton

And that brings us right back to business outcomes. A CFO does not ultimately care if an agent used fewer tokens or executed a clever model routing strategy. The CFO cares whether gross margin improved, whether customer churn decreased, or whether revenue per employee expanded.

Catherine Spencer

Which brings us to our final core theme: governance, authority, and how enterprise leadership must structure the operational control plane for digital labor. Deloitte research recently found that only twenty one percent of organizations believe they possess a mature governance model for agentic AI. And IBM reported that a mere eleven percent of technology executives feel fully prepared for the scale of agent deployment they expect in the next two years.

Edward Hamilton

Those numbers are sobering, but completely understandable. Traditional software governance was designed for static applications with fixed rule sets and human users. Runtime governance for autonomous agents must operate dynamically in real time. You need zero trust agent identities, cryptographic authorization tokens, real time token rate limits, and automated circuit breakers that instantly isolate an agent if its behavior drifts from expected operational parameters.

Charles Skamser

And here is a controversial point I want to put on the table: I believe the primary bottleneck to enterprise AI adoption over the next three years is not going to be model intelligence or context window size. The primary bottleneck is an organization's willingness and legal ability to delegate operational and financial authority to non human actors.

Catherine Spencer

Oh, absolutely Charles! An agent can be infinitely intelligent, but if your corporate policy requires a human manager to manually approve every single sub step, action, and transaction, you have created a massive human bottleneck that completely destroys the speed and economic advantage of digital labor.

Edward Hamilton

It is the shift from a traditional organizational pyramid, where human workers use software tools, to an orchestrated hybrid model, where human managers supervise autonomous digital teams. In this new architecture, human leadership sets objectives, establishes boundary guardrails, defines acceptable risk thresholds, and monitors outcomes, while digital entities handle end to end execution.

Charles Skamser

That is why our work at PX42, with initiatives such as an entity-centric, relationship-derived authority and governed execution platform, Verified Truth, UBIX, and EnterpriseBusinessHealth360, is so critical. You need trusted relationships between entities, verified underlying data, and continuous real-time business health telemetry to make autonomous execution safe and effective.

Catherine Spencer

And we are seeing this exact philosophy applied in specialized vertical domains too, like 360Dealership Health in the automotive sector. Instead of just deploying an isolated chatbot to answer internet leads, you monitor the holistic operational health of the entire dealership group, connecting inventory aging, gross margin leakage, sales conversion, and service capacity into a unified predictive model that guides both human executives and digital agents.

Charles Skamser

Exactly, Catherine. Whether you are managing an automotive dealership group, an enterprise sales forecast in Salesforce using MEDDPICC discipline, or a global banking workflow, the pattern is identical: observe continuously, verify the facts, understand economic impact, predict friction, and execute course corrections in real time.

Edward Hamilton

So Charles, as we bring this episode to a close, if you were to leave our listeners with one decisive diagnostic question to take into their executive boardrooms on Monday morning, what would it be?

Charles Skamser

I would ask every CEO, CFO, and CIO this simple question: Is your leadership team merely buying and managing software tools, or are you actively prepared to govern, orchestrate, and optimize a digital workforce? Because if you are still treating AI as just another software license, the competitive landscape is going to pass you by while you are suspended in airplane mode.

Catherine Spencer

I think that is the definitive question for 2026 and beyond.

Edward Hamilton

A magnificent framing, Charles. And on that note, please enjoy the remainder of your stay in Melbourne, try to get some sleep, and try not to let the International Date Line confuse your calendar too much further!

Charles Skamser

Thank you both! Keep everything under control back home until I return in September. Cheers, everyone!