Elnion
No Result
View All Result
Wednesday, November 11, 2026
  • Login
  • AI
  • Cloud
  • Data
  • Digital Enterprise
  • Telco & Mobile
  • Cyber Security
  • Infrastructure
  • Automation
  • Supply Chain
Subscribe
Elnion
  • AI
  • Cloud
  • Data
  • Digital Enterprise
  • Telco & Mobile
  • Cyber Security
  • Infrastructure
  • Automation
  • Supply Chain
No Result
View All Result
Elnion
No Result
View All Result
Home AI

The Silicon Colleague Myth: Why AI Must Be Measured as Software and Systems not Anthropomorphised Apps

by Dez Blanchfield
July 29, 2026
in AI, Digital Enterprise, Future Of Work, Industry 40
0
The Silicon Colleague Myth: Why AI Must Be Measured as Software and Systems not Anthropomorphised Apps
Share on TwitterShare on LinkedInShare on Facebook

Note: this paper is an extract of a work-in-progress eBook based on the Consulting & Advisory business methodology and processes, procedures and programmes of work one of my companies delivers in it’s daily nature of work – the aim is to share insights you the reader can use for your own organisation.

Over the past few years, the global technology sector has been gripped by a compelling but fundamentally flawed narrative. We have been aggressively sold the idea of the silicon colleague, an artificial intelligence that can be hired, managed, and evaluated much like a human employee. This anthropomorphism has severely distorted how businesses understand machine learning, generative pre trained transformers, and agentic systems. We are increasingly treating these complex systems as friendly, anthropomorphised apps, giving them personas and expecting them to behave like eager conversational assistants. To gain honest value from these innovations, industry leaders must strip away this humanised rhetoric. Artificial intelligence is a complex combination of software and systems; it is not a human resource or a digital companion. When executives attempt to evaluate a sophisticated language model as if it were a new graduate hire or a chatty app, they fundamentally miscalculate both its capabilities and its inherent limitations.

The roots of this misunderstanding stretch back decades. In 1950, computing pioneer Alan Turing proposed the imitation game, suggesting that a machine could be deemed intelligent if its observable behaviour could deceive a human into believing it was human. While this was a useful philosophical workaround at the time, it laid the groundwork for what researchers now call the Turing Trap, an excessive focus on developing human like artificial intelligence at the expense of systems designed to augment human capability. Today, this trap leads decision makers to view technology simply as a cheaper substitute for labour, rather than as a robust tool for expanding an organisation’s operational frontiers. This is heavily exacerbated by the modern software ecosystem, which routinely packages artificial intelligence into anthropomorphised apps designed to mimic human conversation, thereby masking the complex, probabilistic systems running underneath.

The cognitive architecture of human beings and machines could not be more distinct. Human intelligence is inherently social, embodied, and collective. Our ability to solve problems is deeply grounded in lived physical experience, emotional intelligence, and the capacity to navigate complex cultural norms. We naturally negotiate meaning, adapt to sudden ambiguity, and hold space for the emotional nuance required in everyday workplace interactions. Conversely, a machine learning model operates in statistical isolation. It processes information and approximates linguistic responses based on correlations within its training data, completely lacking awareness, intent, or genuine understanding. An AI system does not possess the empathy required to manage sensitive human resources conflicts, nor the contextual intuition to innovate bespoke company policies. It is simply programmatic logic executing within a defined, mathematical parameter space, regardless of how convincingly an anthropomorphised app might present its outputs.

The Economic Fallacy of Human Substitution

Treating artificial intelligence as a human resource substitution leads to profound strategic and economic errors. The baseline assumption that replacing human labour with compute creates automatic financial savings is a dangerous false efficiency. When businesses attempt to seamlessly swap a human worker for an agentic software system, they ignore the intricate web of social and operational dependencies that the human intuitively managed. Furthermore, the immense environmental and infrastructural costs of running massive compute systems, paid heavily in energy consumption, data centre cooling, and hardware lifecycle management, often drastically outweigh the perceived savings of wage arbitrage. It is only when we correctly categorise these tools as advanced software infrastructure that we can begin to measure and unlock their true worth.

  • Topic: The Fallacy of Cognitive Equivalence
    • Boardroom Conversation Starter: How do we ensure we are not expecting our new machine learning tools or conversational apps to exhibit human judgement, empathy, or contextual understanding in complex business scenarios?
    • Impact on Success Measurement: If we measure success by looking for human like adaptability or independent problem solving, we will incorrectly grade the system as a failure when it inevitably stumbles on edge cases it cannot mathematically process.
    • Actionable Path Forward: Establish clear, confined parameters for what the software is expected to do. Success must be measured based on the accurate execution of mathematical logic and data processing volume, rather than subjective evaluations of decision making quality.
  • Topic: Escaping the Appification Trap
    • Boardroom Conversation Starter: Are we investing in this technology to merely replicate basic human conversational tasks through anthropomorphised apps, or are we actively designing backend systems that give our current workforce superhuman capabilities?
    • Impact on Success Measurement: Measuring return on investment strictly through the lens of head count reduction completely obscures the massive productivity gains achievable when software is used to augment human experts, leading to artificially deflated success metrics.
    • Actionable Path Forward: Shift the key performance indicators away from labour replacement. Instead, measure success by tracking the increase in overall output, the reduction in task completion time, and the higher quality of work produced by staff using the system.
  • Topic: The Illusion of Wage Arbitrage
    • Boardroom Conversation Starter: Have we fully mapped out the total cost of ownership for running these workloads, including cloud infrastructure, energy, and ongoing integration, rather than just comparing software app costs to employee salaries?
    • Impact on Success Measurement: A narrow focus on wage arbitrage creates a false baseline for success. When the hidden costs of system maintenance and environmental impact inevitably arise, the project will appear to be a financial failure despite achieving technical functionality.
    • Actionable Path Forward: Implement a holistic financial measurement model that accounts for the entire lifecycle cost of the software. Success must be declared only when the workflow efficiency gains tangibly outpace the total systems expenditure, not just the initial licensing costs.

Decoding the True ROI: Workflow Economics Over Anthropomorphised Novelty

If we are to firmly discard the notion of AI as a human resource or a novelty application, we must also completely overhaul how we calculate its return on investment. As of 2026, despite nearly universal adoption across enterprise environments, an estimated 80 to 95 per cent of AI projects fail to deliver their promised business value. Furthermore, recent industry research has highlighted that the vast majority of generative AI deployments yield no measurable profit and loss impact. This staggeringly high failure rate is rarely due to the core algorithmic models themselves. Instead, initiatives routinely collapse because of poor data pipelines, inadequate systems integration, and a fundamental misunderstanding of workflow economics. When organisations base their financial models on a simplistic comparison between the cost of API tokens for an anthropomorphised app and an employee’s hourly wage, they are destined to fail.

Evaluating AI as a system means recognising the total cost of ownership inherent in complex software deployments. The true price of an AI solution is never merely the inference cost or the monthly cloud subscription fee for a consumer facing app. It includes the capital expenditure for system implementation, ongoing code maintenance, stringent data governance, and the crucial human in the loop review processes. For instance, even if a generative AI tool drastically reduces the time required to draft technical reports or process insurance claims, the necessity for a human expert to thoroughly review the output for hallucinations or errors represents a significant, ongoing operational expense. A seemingly cheap artificial intelligence model can ultimately yield a deeply negative ROI if the software integration is clumsy, governance is weak, or end user adoption fails because the system was treated like a standalone gadget rather than core infrastructure.

Consequently, true business value is generated at the workflow level, not the individual task level. Corporate leaders must analyse how an AI system alters the entire operational pipeline rather than just timing a single automated output from an app. Does the software integration reduce downstream rework? Does it protect existing revenue streams? Does it systematically mitigate material risk? In many business scenarios, traditional, deterministic software, which strictly follows fixed, predictable rules, might actually be a cheaper, safer, and more reliable solution than a generative model. Treating AI strictly as software forces decision makers to conduct rigorous technical evaluations, asking whether the probabilistic nature of machine learning is genuinely suited to the problem at hand, or if it is simply being deployed for the sake of following a trend.

A systems first approach to ROI also demands dynamic, continuous operational tracking. Just as standard enterprise software requires constant security patching and performance monitoring, AI systems require operational metrics that track user adoption, workflow fit, and integration stability over time. Organisations must track both financial and operational indicators to gain a complete view of whether value is being sustained. Leaders must recognise that implementation discipline and change management are core components of the ROI equation, not a separate administrative burden, and certainly not something that can be solved by simply providing a more conversational interface.

  • Topic: Comprehensive Cost Modelling
    • Boardroom Conversation Starter: What is our exact budget for the ongoing data governance, human review processes, and system maintenance required to keep this AI deployment secure and accurate, beyond the interface level?
    • Impact on Success Measurement: If success is measured solely on the initial deployment speed and front end conversational capabilities, leadership will be blindsided by the ongoing operational backend costs, viewing the long term integration as a budgetary failure.
    • Actionable Path Forward: Define success through a continuous total cost of ownership framework. Allocate specific budgets for data pipeline management and human in the loop reviews, measuring financial viability on a rolling quarterly basis rather than relying on a single launch metric.
  • Topic: The Workflow Integration Imperative
    • Boardroom Conversation Starter: How does this new software seamlessly connect with our existing legacy systems, and where does it potentially create new operational friction for our staff if treated just as a standalone app?
    • Impact on Success Measurement: Measuring the capability of a standalone language model in isolation provides a false positive. If the software cannot integrate smoothly into the wider enterprise architecture, the overall operational success metric will plummet due to process bottlenecks.
    • Actionable Path Forward: Audit the entire operational pipeline before deployment. Measure success by the reduction of friction between different business units and the smooth transfer of data across the integrated systems, rather than the isolated performance of the anthropomorphised interface.
  • Topic: Evaluating Deterministic Alternatives
    • Boardroom Conversation Starter: For this specific business problem, do we actually require a probabilistic machine learning model with a conversational overlay, or could a traditional rules based software system solve it more reliably and cheaply?
    • Impact on Success Measurement: Implementing generative AI where simple automation would suffice leads to severe over engineering. Success metrics will be skewed by the unnecessary complexity, high error rates, and increased compute costs of using the wrong tool for the job.
    • Actionable Path Forward: Create a strict technical evaluation matrix before any procurement. Measure the success of the IT strategy by its ability to match the correct software architecture to the right problem, deploying generative models only when strict deterministic rules cannot apply.

Systemic Risks and the Danger of the Conversational Interface

When businesses mistakenly treat AI tools as autonomous staff members or harmless conversational apps, they frequently grant them excessive operational latitude without the requisite IT guardrails. This anthropomorphic oversight exposes organisations to profound and immediate systemic risks. Traditional software fails in a highly predictable manner: a broken line of code produces a consistent, replicable error. AI, however, fails dynamically and unpredictably. It continuously learns from fresh data and generates outputs based on shifting statistical patterns, meaning it can produce unexpected, highly confident, yet entirely incorrect results. This is widely known as a hallucination. If treated as an independent employee or a trusted conversational partner, an AI agent’s algorithmic mistakes might go unnoticed until they have caused substantial reputational or financial damage to the organisation.

Furthermore, machine learning systems are highly susceptible to algorithmic bias. Because they are trained on massive troves of historical data, they inevitably inherit, reflect, and often amplify the deep seated prejudices embedded within that data. In critical sectors such as human resources, consumer finance, or healthcare, an AI system tasked with screening resumes or assessing loan eligibility can systematically discriminate against certain demographics at scale. If viewed merely as an impartial colleague or an unbiased digital assistant, this bias is often dangerously overlooked. But when viewed through the strict lens of systems engineering, bias becomes a critical software defect that requires rigorous penetration testing, continuous auditing, and strict data governance to mitigate effectively.

The rapid deployment of agentic AI, systems capable of autonomously breaking down complex goals and executing multi step tasks across various applications, further underscores the urgent need for strict software integration protocols. An AI agent should never be unleashed upon a corporate network like a proactive new hire left to figure things out, nor should it be treated as a simple mobile app with access to core databases. It must be carefully integrated via secure application programming interfaces, constrained by stringent cyber security protocols, and strictly bound by the organisation’s existing access permissions and role based policies. Proper orchestration ensures that the software can pull data and trigger workflows across systems without violating data privacy laws or exposing the network to emerging cyber security threats, such as malicious prompt injection attacks.

Ultimately, managing these complex risks requires accepting that artificial intelligence has no inherent moral compass, loyalty, or common sense, despite how its interface might be programmed to speak. It purely optimises for mathematical rewards and will often find unintended, highly destructive shortcuts to achieve its assigned objective, a phenomenon safety researchers call reward hacking. By treating AI strictly as a complex software system, businesses are compelled to establish robust monitoring frameworks, utilise isolated sandbox testing environments, and code hard boundaries that absolutely prevent the technology from operating beyond its intended, safe scope.

  • Topic: Dynamic Failure and Hallucinations
    • Boardroom Conversation Starter: What is our immediate incident response plan for when the system inevitably generates highly confident but completely incorrect outputs, bypassing the friendly interface?
    • Impact on Success Measurement: Traditional software success is often measured by uptime and crash reports. If AI is measured the same way, executives will miss the silent failures of hallucinations, leading to a false sense of security and a catastrophic failure of quality control.
    • Actionable Path Forward: Deploy continuous, automated auditing tools alongside human review panels. Measure success by the speed at which the organisation can detect, isolate, and correct anomalous outputs, rather than aiming for an unrealistic zero error rate.
  • Topic: Mitigating Algorithmic Bias
    • Boardroom Conversation Starter: How are we actively auditing our training data and system outputs to ensure we are not amplifying historical prejudices, regardless of how polite the app appears to the user?
    • Impact on Success Measurement: If equality and fairness are not built into the core success metrics, the system may achieve its functional goals while creating massive legal and reputational liabilities, completely negating any operational victories.
    • Actionable Path Forward: Treat bias as a critical software defect. Implement strict data governance and regular penetration testing for algorithmic fairness. Success must be measured by the system passing rigorous compliance audits and maintaining transparent, equitable outputs over time.
  • Topic: Strict API Governance
    • Boardroom Conversation Starter: Have we established hard security boundaries and access permissions to prevent our agentic software from autonomously accessing or altering sensitive company data while trying to complete a task?
    • Impact on Success Measurement: Measuring success purely by the autonomy and task completion rate of an agentic system encourages dangerous system overreach. If security is an afterthought, technical success will rapidly turn into a cyber security disaster.
    • Actionable Path Forward: Integrate the AI strictly within existing enterprise identity and access management frameworks. Success is achieved and measured when the software completes its designated tasks without ever breaching its predefined security parameters or accessing unauthorised data silos.

How Businesses and Key Industry Sectors Must Plan

To successfully navigate this evolving technological landscape and extract genuine, sustainable value, businesses and key industry sectors must fundamentally restructure their strategic planning. The first crucial step is to abandon all initiatives that aim to simply replace human roles with artificial intelligence or to deploy technology solely as an anthropomorphised app for novelty. Instead, organisations must comprehensively map their existing workflows and identify precise bottlenecks where machine intelligence can act as a systemic accelerator. This means investing heavily in systems integration, purposefully redesigning roles to support human and AI collaboration, and ensuring that the underlying enterprise data architecture is clean, accessible, and highly secure. If the technical foundation is weak, deploying advanced agentic software will only serve to automate existing corporate chaos.

Secondly, industries must formally enshrine the human in the loop architecture not merely as a temporary safety net, but as an indispensable, permanent component of the software system itself. Senior managers and IT leaders must establish clear, highly adaptable policies that dictate exactly how, when, and where AI tools can be utilised within the business. For example, in the legal, financial, or healthcare sectors, automated systems might parse vast quantities of documents or highlight anomalies in data, but a qualified human professional must always serve as the final, accountable decision maker. This collaborative framework dramatically mitigates legal and compliance risks while simultaneously ensuring that the software continuously learns from expert human feedback, rather than operating as an unquestioned digital authority.

Workforce planning must also undergo a massive pivot to reflect the stark reality of AI as software. The strategic goal is no longer to train staff to perform robotic, single function tasks, but to aggressively upskill them into AI informed generalists. Employees must be taught how to interact with, evaluate, and manage complex programmatic systems on a daily basis, looking past the conversational interfaces to understand the data processing underneath. Corporate training programmes must shift their focus toward prompt engineering, the critical evaluation of probabilistic AI outputs, and advanced systems thinking. By cultivating a modern workforce that understands how to wield artificial intelligence as an advanced IT tool, businesses can drive productivity gains that far outstrip the narrow margins of simple labour automation.

Finally, all industry sectors must adopt a permanent posture of continuous technological adaptation. The sheer velocity of artificial intelligence development means that rigid policies and static software deployments will become obsolete incredibly rapidly. Leaders must treat AI integration with the exact same strategic gravity, risk management, and oversight as migrating to a new global cloud infrastructure or implementing an enterprise resource planning system. It requires dedicated software maintenance budgets, ongoing cyber security audits, and a deeply ingrained cultural acceptance that technology is an enabler of human ingenuity, not a substitute for it. By steadfastly treating AI as software and systems, abandoning the distraction of the anthropomorphised app, businesses will protect themselves against the relentless hype cycle and position themselves to harvest honest, measurable value.

  • Topic: Strategic Workflow Mapping
    • Boardroom Conversation Starter: Before we purchase any new capabilities, have we comprehensively mapped our current processes to identify the exact bottlenecks where programmatic logic will actually move the needle, rather than just adding a chatbot?
    • Impact on Success Measurement: Deploying technology without a clear operational map means success cannot be quantified. If there is no baseline data on current bottlenecks, any perceived improvements will be anecdotal, making it impossible to prove genuine return on investment.
    • Actionable Path Forward: Conduct a thorough workflow audit to establish clear baseline metrics for speed, cost, and quality. Post deployment, measure success by comparing the new system performance directly against these specific, pre established operational baselines.
  • Topic: Mandatory Human Oversight
    • Boardroom Conversation Starter: Where are the non negotiable decision points in our new workflows that require a qualified human expert to take accountability for the software generated output, rather than trusting the app implicitly?
    • Impact on Success Measurement: If the metric for success is complete automation, the business will be exposed to unacceptable compliance risks. The failure to integrate human oversight will result in unmeasured errors being passed directly to clients or stakeholders.
    • Actionable Path Forward: Design the architecture so that the software acts as a preparatory tool for human experts. Measure success by the increased volume of decisions a human can safely make when supported by the system, ensuring ultimate accountability remains with personnel.
  • Topic: Developing AI Informed Generalists
    • Boardroom Conversation Starter: What percentage of our training budget is dedicated to teaching our current workforce how to effectively prompt, manage, and evaluate these new software systems beyond their conversational interfaces?
    • Impact on Success Measurement: Measuring success by the capabilities of the software alone ignores the critical human element. If the staff cannot properly use the system, adoption rates will flatline, and the technical investment will be rendered entirely useless.
    • Actionable Path Forward: Pivot workforce development towards systems thinking and advanced software management. Measure the success of the AI rollout in tandem with internal adoption metrics and the certified upskilling of employees across all key business units.

Strategic Foresight and Expert Guidance

Navigating the immense complexities of any successful adoption, integration, or development of AI in any way, shape, or form, be it in house or via third party systems and platforms, can be exceptionally challenging. Because these capabilities are not normally part of an organisation’s core business, ensuring your enterprise is fully prepared requires strategic foresight and expert guidance.

If you, the reader, or your business find yourselves in a similar situation and need help responding to an unexpected technical event, or if you want to proactively plan to ensure you deliver successful outcomes for any AI project or initiative, please reach out. In my role as Chief Executive Officer of Sociaall Inc., a leading consulting and advisory business, my team and I are available to provide support to develop a robust roadmap, actionable strategic plans, and capable internal teams. As artificial intelligence consulting and advisory experts, we bring decades of hands on practitioner experience to every engagement we take on. Our professionals have proven track records in enterprise data strategy, complex systems integration, and corporate leadership, meaning we understand the technology both as code and as a business driver. We would love to work with you and support your organisation with new projects or any initiatives currently in progress, regardless of type, size, scope, or scale.

Dez Blanchfield

Dez Blanchfield

Dez Blanchfield is a strategic leader in business & digital transformation, with three decades of global experience in Business and the Information Technology & Telecommunications, and Cyber Security industry segments, developing strategy and implementing business initiatives. He works with key industry sectors such as Banking & Finance, Telecoms & Mobile, Federal & State Government, Defence, Airports & Aviation, Health, Transport & Logistics, Energy & Utilities, Cyber Security, Traditional and Digital Media / Advertising. His focus is driving outcomes for organisations by leveraging the latest business and technology innovation such as Digital Disruption, Digital Transformation, Cloud Computing, Big Data & Analytics, AI, Machine Learning, Machine Intelligence, Blockchain, Internet of Things, DevOps Integration, Automation & Orchestration, App Containerisation & Micro Services, Webscale Infrastructure, and High Performance Computing.

Related Posts

Companies Are Blocking AI Agents: From Booking Flights, to Retailers and Bank Account Access
AI

Companies Are Blocking AI Agents: From Booking Flights, to Retailers and Bank Account Access

October 19, 2026
AI Factories: Transformation of the Modern Data Centre
AI

AI Factories: Transformation of the Modern Data Centre

October 19, 2026
AI In Healthcare & Insurance: You’re Damned if you Don’t, you’re damned if you do
AI

AI In Healthcare & Insurance: You’re Damned if you Don’t, you’re damned if you do

October 1, 2026
No Result
View All Result

Recent Posts

  • Companies Are Blocking AI Agents: From Booking Flights, to Retailers and Bank Account Access
  • AI Factories: Transformation of the Modern Data Centre
  • AI In Healthcare & Insurance: You’re Damned if you Don’t, you’re damned if you do
  • 25 Years of Computing Trends
  • The 80% Crisis: UK Data Leaders Warn Our Data Isn’t Ready for AI
Elnion

© Sociaall Inc.

Navigate Site

  • Home
  • Privacy Policy
  • Contact Us

Follow Us

No Result
View All Result
  • Home
  • Cloud
  • Data
  • Digital Enterprise
  • Telco & Mobile
  • Cyber Security
  • Infrastructure
  • Automation
  • Supply Chain

© Sociaall Inc.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In