The Silo Surcharge: How Disconnected Tools Bleed Productivity

Many organizations believe they have digitized because they use software for payroll, a different tool for learning, and yet another for performance reviews. However, a collection of disconnected tools does not constitute a digital ecosystem; it creates a fragmented landscape that bleeds productivity. This fragmentation imposes a “Silo Surcharge” on every interaction, primarily through the phenomenon known as context switching.

The Swivel Chair Effect

Context switching-the act of toggling between disparate applications-is a major productivity destroyer. Research indicates that employees can lose up to 40% of their productive time simply by shifting mental gears between tasks and tools. In an HR context, this “swivel chair” effect occurs when a manager must log into an LMS to assign training, a separate HRIS to check a job description, and a spreadsheet to record a performance note. This fragmentation breaks the flow of work, turning simple talent management tasks into administrative burdens that managers actively avoid.

Data Integrity Debt and Analytics Blindness

The secondary cost of silos is data blindness. When performance data lives in one island and learning data in another, it becomes impossible to correlate training investments with performance outcomes. Organizations accumulate “Data Integrity Debt” as they rely on manual exports and Excel gymnastics to reconcile conflicting datasets. This prevents leadership from making evidence-based decisions. You cannot optimize what you cannot see. Without a unified view of the talent journey, strategic planning becomes a guessing game, leaving the organization vulnerable to market shifts that a data-integrated competitor would have anticipated months in advance.

The IT Maintenance Burden

Finally, the cost of maintaining custom bridges between legacy systems is often underestimated. IT departments spend disproportionate resources building and patching APIs or managing secure file transfers between incompatible vendors. This maintenance budget yields no new value; it merely keeps the lights on. By consolidating these functions into an integrated platform, organizations eliminate the technical debt of fragmentation and redirect IT resources toward value-generating initiatives.

The Manual Labor Debt: Paying Human Rates for Robot Work

Perhaps the most irrational expense in the CODN calculation is the “Manual Labor Debt”-the practice of paying skilled human professionals to perform repetitive, administrative tasks that could be automated. When HR leaders spend the majority of their week chasing approvals, manually updating spreadsheets, or answering routine policy questions, they are trapped in low-value transactional work. This prevents them from engaging in the strategic initiatives-such as succession planning and culture building-that actually drive business growth.

The High Cost of Human Error

Manual processes are not only slow; they are prone to error. A single data entry mistake in compliance reporting or payroll can result in fines that far exceed the cost of automation software. Furthermore, the speed of service delivery impacts the employee experience. In a manual environment, an employee request might sit in an inbox for days. In an automated environment, workflows trigger immediate actions. According to McKinsey & Company, automation can reduce the time spent on typical HR administrative tasks by up to 50%, effectively doubling the capacity of the existing team without adding headcount.

Gamification as Automation

Automation extends beyond simple workflows; it includes behavioral nudges. Modern platforms use gamification mechanics to automate engagement. Instead of HR manually reminding employees to complete training, the system uses automated progress bars, badges, and peer leaderboards to drive participation intrinsically. This “set it and forget it” approach to engagement ensures consistent execution of talent programs without requiring constant manual intervention. The scalability issue is the final nail in the coffin for manual processes: as a company grows, manual work grows linearly, eventually breaking the system. Automation is the only way to decouple growth from administrative overhead.

Read more:

• Mastering Employee Turnover and Retention
• Navigating the Top HR Trends of 2025
• Understanding Job Hopping Trends

In this article:

Anyone who has worked in a company with more than three departments knows what it is to send an email asking for data and wait two days for a reply that, when it finally arrives, is already out of date. Now multiply that problem by hundreds of processes, thousands of employees and dozens of systems that do not talk to one another. This is the daily reality of most organizations.

AI agents promise to solve this fragmentation. But an isolated agent – however intelligent it may be – has the same limitations as a brilliant employee locked in an office with no phone and no network access. The real leap happens when those agents can communicate with one another, access data from multiple sources and coordinate actions in real time. And that is only possible thanks to a new generation of communication protocols.

For technology leaders, understanding how these protocols work is not an academic exercise. It is the difference between deploying AI agents that genuinely transform operations and investing in technology that ends up creating yet another isolated island within the company.

The Evolution of Interoperability: From APIs to Agent Protocols

Before APIs became commonplace, integrating two systems was a heavy engineering project. Every connection between applications required custom development, constant maintenance and an enormous amount of patience. APIs came along to standardize that communication: they created a common language so that different systems could exchange data in a structured way.

Thanks to APIs, the digital ecosystem flourished. Your ERP began to communicate with the billing system, the CRM automatically feeds the sales dashboards, and the e-commerce platform syncs stock with the warehouse. All of this works because there is a clear contract – an API – between each pair of systems.

But APIs were designed for a world in which humans decide and systems execute. The request always comes from a person or a programmed process. Now we are entering a different reality: AI agents need to make autonomous decisions, access context in real time and coordinate with other agents without constant human intervention. Traditional APIs, while fundamental, were not conceived for this kind of dynamic, context-aware interaction.

This is where agent communication protocols come in – the next chapter of enterprise interoperability.
At GFoundry, we spent years building integrations with ERPs, HR systems and some of our clients’ proprietary platforms. We know the complexity of connecting systems well. But what we are seeing now is a qualitative leap: it is no longer just about exchanging data between platforms, but about enabling intelligent agents to navigate multiple systems to solve talent management problems that previously took days of manual work. João Carvalho – CEO, GFoundry

The Three Foundational Protocols: MCP, ACP and A2A

The agent communication ecosystem is consolidating around three main protocols. Each one solves a different problem, and it is in their combination that the real potential lies.

Model Context Protocol (MCP): The Protocol That Gives Agents Context

An AI agent with no access to relevant data is like a consultant who arrives at a company without knowing the sector, the customers or the numbers. They may be highly capable, but they will not produce anything useful without context.

MCP, created by Anthropic as an open protocol, solves precisely this problem. It defines the way applications based on Large Language Models (LLMs) connect to data sources and external tools. Instead of every developer having to build custom connectors for each database or service, MCP standardizes that connection.

In practice, MCP allows an AI agent to access internal files, query databases, invoke specific tools and incorporate external data – all in a standardized way and in real time.

A concrete example: imagine an AI agent that assists team managers within a talent management platform. When a manager asks “which employees on my team completed less than 50% of their training plan this quarter?”, the agent uses MCP to query the Learning & Development module, cross-references it with attendance data on another platform and delivers a context-aware answer – all in seconds, without the manager having to navigate three different systems.

This is exactly the scenario GFoundry delivers with its MCP approach, brought together in the TOAR model (Talent Orchestration, Automation & Response): the platform exposes its capabilities through MCP, so that you can manage talent, measure metrics, extract and process data and obtain recommendations from Microsoft Copilot, ChatGPT or Claude, and cross talent data with other systems that also speak MCP, without ever entering the back office. See how GFoundry’s TOAR model works.

Agent Communication Protocol (ACP): Coordination in a Local Environment

If MCP is about connecting agents to data and tools, ACP focuses on how multiple agents communicate and coordinate within the same environment.

Think of ACP as the internal protocol of a well-organized project team. Each member knows what the others are doing, what skills each one has, and how to divide the work efficiently. ACP creates exactly this kind of coordination for AI agents operating within the same platform or infrastructure.

ACP standardizes the way agents share skills, distribute tasks and communicate the status of their activities. It removes the need for custom interfaces between every pair of agents, creating a space for fluid collaboration.

A concrete example: within a talent management platform like GFoundry, several agents can operate in parallel – one focused on recruitment, another on engagement, another on performance analysis. When the performance agent detects that a team has consistently below-average evaluations, ACP allows it to communicate directly with the training agent to suggest relevant development content, with no need for human intervention to bridge the two.

Agent-to-Agent (A2A): Collaboration Across Platforms and Vendors

MCP connects agents to data. ACP coordinates agents within a system. But what happens when agents from different systems and vendors need to work together?

The A2A protocol, introduced by Google, solves this question. It defines how agents from different platforms can initiate tasks with one another, share information, send real-time updates and exchange files – all in an interoperable way.

This is particularly important in the enterprise context, where a single platform rarely covers every need. The technology ecosystem of any mid-sized company includes dozens of tools from different vendors.

A concrete example: consider the process of an employee leaving the company. An agent on the HR platform detects the departure and uses A2A to communicate with an agent in the IT system (which initiates access revocation), with an agent in the finance system (which processes the final payments) and with an agent on the knowledge management platform (which schedules know-how transfer sessions). Each agent operates in a different system, from a different vendor, but A2A ensures they collaborate as if they were a single team.

In talent management, no platform operates in isolation. Our clients use ERPs, payroll systems, internal communication tools, external LMSs – all at the same time. The promise of A2A is that AI agents can orchestrate processes that span all of these tools. For a platform like GFoundry, this means that our engagement or training agents can collaborate natively with agents from our clients’ other systems. It is a paradigm shift. João Carvalho – CEO, GFoundry

How the three protocols complement each other

None of these protocols replaces the others. They work in complementary layers:

Strategic Implications for CIOs and CTOs

Understanding these protocols is not a technical luxury – it is a strategic necessity. For those making decisions about technology infrastructure, the implications are profound and immediate.

End-to-end automation that actually works

Most attempts at enterprise automation run into the same problem: processes cut across multiple systems and departments, and automation only works within each silo. These protocols remove that barrier.

When agents can access context-aware data (MCP), coordinate within a platform (ACP) and collaborate across systems from different vendors (A2A), it becomes possible to automate complete processes – not just fragments.

Example: the onboarding process for a new employee typically involves HR, IT, training, compliance and the team manager themselves. Today, this process relies on emails, manual checklists and a lot of goodwill. With agents orchestrated by these protocols, the process can run autonomously: from preparing the equipment to automatic enrolment in mandatory training modules, including the creation of access credentials and a personalized introduction to the company culture – all without anyone having to send a single email.

Decisions based on real data, not on hunches

One of the biggest challenges in enterprise decision-making is that the relevant information is scattered across multiple systems, often outdated and hard to cross-reference. MCP changes this reality by allowing AI agents to query real-time data from multiple sources and present it in a context-aware way.

Instead of waiting for quarterly reports to realize there is a retention problem on a particular team, an agent can cross-reference engagement data, performance evaluations and absenteeism patterns to flag the risk before the employee even starts looking for alternatives.

Human oversight: the guarantor of trust

Agent autonomy raises a legitimate question: who controls what they do? These protocols were designed with transparency built in. Agents’ actions are traceable, auditable and, at critical points, require human approval.

This “human in the loop” approach is not a limitation – it is an essential feature. It allows organizations to benefit from the speed and scale of agents without giving up control over sensitive decisions. An agent can prepare the entire analysis and recommendation, but the final decision on a promotion, a restructuring or a response to an important client remains human.

Cybersecurity: the new perimeter to defend

Agents with access to sensitive data represent a new vector of risk. If an agent can access performance information, salary data or confidential files, then compromising that agent is equivalent to compromising all of that information.

Both MCP and A2A incorporate security mechanisms – host-mediated access control, sandboxing, OAuth 2.0 and API key authorization. But technology alone is not enough. CIOs and CTOs need to define clear policies on which data each agent can access, which actions it can execute autonomously and which decisions require human validation.

There is a natural temptation to focus all attention on agents’ capabilities – on what they can do. But the more important question for a CIO or CTO is: what should they not do? Defining the limits is as critical as defining the capabilities. At GFoundry, we handle training data, performance evaluations, career paths, 360 feedback – information that, if misused, can have serious consequences. Governance has to be designed at the same level as functionality. João Carvalho – CEO, GFoundry

How to Build Your AI Agent Strategy

Theory matters, but what counts is execution. Here are practical guidelines for leaders who want to integrate these protocols into their organization’s AI strategy.

1. Start with a real process, not an abstract concept. Choose a concrete business process that is complex enough to benefit from multiple agents, but bounded enough to allow a controlled implementation. Employee onboarding, application management or the performance evaluation cycle are good starting points. Identify where the bottlenecks are, which systems are involved and which decisions could be accelerated by agents with access to context-aware data via MCP.

2. Map out the interactions between agents before building. Before implementing, clearly define which agents will exist, what skills each one will have and how they will communicate with one another. If all the agents operate within the same platform, ACP is the relevant protocol. If the process involves systems from different vendors – which is almost always the case – A2A will be necessary. Also think about the data that will be shared: what information each agent needs, what information it should not access and how to ensure that interactions between agents do not create vulnerabilities.

3. Build in security and governance from day zero. Do not treat security as a phase that comes after implementation. Define access policies before giving data to the agents. Establish human approval points for sensitive decisions. Implement logging and auditing of all agent actions. And create “circuit breaker” mechanisms that allow an agent to be shut down or limited quickly if something goes wrong.

4. Invest in interoperability and open standards. Avoid getting locked in to a single vendor. By adopting platforms and tools that support open protocols such as MCP and A2A, you ensure that your agents can evolve and integrate with the wider ecosystem. Middleware – software that acts as a universal translator between different agents and systems – will become increasingly important. Bet on solutions that include it or that make its implementation easier.

5. Measure, learn and iterate. Define clear metrics from the outset: time saved, errors avoided, user satisfaction, cases where a human had to intervene. Use this data to continuously refine the strategy, expand to new processes and adjust agents’ permissions and capabilities.

The most common mistake I see in companies is wanting to implement AI in everything at the same time. My recommendation is the opposite: choose a process where the pain is real and measurable, implement agents in a controlled way, measure the results and only then scale. At GFoundry, we started by applying agents to support the creation of training content and to analyze engagement patterns. The results in those specific processes gave us the confidence to expand. It is a marathon, not a sprint. João Carvalho – CEO, GFoundry

The Future: Unlocking Collaborative Intelligence

The history of enterprise technology is a history of progressive connections. First we connected computers in local networks. Then we connected networks to the internet. APIs connected applications to one another. Now, agent protocols are connecting artificial intelligences to one another.

The rise of MCP, ACP and A2A shows that the industry is converging towards standardization. This is a positive sign: it means we are moving out of the phase of isolated experiments and into a phase of shared infrastructure. The natural concern about fragmentation – too many protocols, too many incompatibilities – is legitimate, but the clear trend is towards convergence.

The real potential does not lie in any individual agent, however sophisticated it may be. It lies in collaborative intelligence – in the ability of multiple agents, from multiple platforms, to work together to solve problems that none of them could solve alone.

For technology leaders, the time to act is now. You do not need to adopt everything at once, but it is essential to understand the terrain, start experimenting and ensure that your organization’s infrastructure is ready for this new reality.

The protocols already exist. The tools are maturing. What is missing is informed leadership to turn potential into concrete results.

We are facing one of those rare windows in which technology advances faster than most organizations can keep up – and that is precisely why there is an enormous opportunity for those who move now. The companies that master AI agent orchestration will not only have more efficient processes; they will have a capacity for adaptation and innovation that the rest simply will not be able to match. At GFoundry, this is the future we are building for. João Carvalho – CEO, GFoundry

The Vital Integration: Correlating Performance (OKRs), Learning and Engagement

True organisational health cannot be measured by a single indicator. The power of the GFoundry platform lies in its ability to cross data that would traditionally be isolated, revealing correlations that escape surface-level analysis. By integrating objective management (OKRs) with the learning management system (LMS) and gamification metrics, we gain a holistic view of the employee.

Leading vs. Lagging Indicators

Most financial metrics are lagging indicators – they confirm what has already happened. By contrast, engagement data in GFoundry works as a leading indicator. A sudden drop in participation in gamified missions or in login frequency on the mobile app often serves as a thermometer of demotivation, visible weeks or months before it shows up in a negative performance review or a resignation letter.

Anomaly Detection and Correlations

Cross-analysis makes it possible to identify complex scenarios:

• High Performance, Low Engagement: Employees who hit their OKRs but do not take part in the company culture (high risk of departure or silent burnout).
• High Engagement, Low Performance: Employees who are very active on the platform and socially integrated, but who fall short on results (an urgent need for training or reskilling).
• Training-Result Correlation: Checking whether the teams with the best scores in e-learning modules are, in fact, exceeding their sales or efficiency targets.

In addition, continuous feedback gathered through pulse surveys and reactions in the internal community turns qualitative data into quantitative sentiment-analysis trends, allowing HR to act on the organisational climate in real time.

Closing Skills Gaps: Data That Guides Reskilling

In an era where the useful life of a technical skill is ever shorter, static skills management has become obsolete. Skills matrices filled in annually in Excel can no longer keep pace with the speed of the market. GFoundry proposes a dynamic approach, where learning and performance data continuously feed the employee’s profile.

Dynamic skills mapping uses artificial intelligence to suggest personalised learning paths. If an employee shows an interest in leadership through voluntary participation in mentoring projects on the platform, the system can automatically recommend an e-learning path on team management. More importantly, the platform makes it possible to identify “hidden talents” – skills that employees hold and demonstrate in gamified challenges, but which do not appear in their official job description.

This data-driven approach makes it possible to calculate the true ROI of training. Instead of measuring only training hours (an effort metric), organisations can measure the effective acquisition of skills and its impact on subsequent business KPIs.

Conclusion: The Future Is Now – Turning Numbers into Success Stories

The era of talent management based on instinct alone has come to an end. The organisations that thrive in today’s competitive landscape are those that can turn cold numbers into stories of human success. Data, on its own, is inert; its value lies entirely in the behavioural or structural change it manages to spark. Advanced People Analytics serves not only to predict who will leave, but to create the conditions for the best people to want to stay.

Technology plays a central role in this transformation, not as a mere repository of records, but as an engine of organisational intelligence. Tools that natively integrate the different dimensions of the employee experience offer an unmatched advantage: the ability to see the whole picture. The next step for HR leaders is to audit their current infrastructure and ask whether it gives them reports to file away or intelligence to act on.

The transition from reactive management to a strategic, data-driven culture demands the right infrastructure. GFoundry operationalises this intelligence, allowing companies such as DPD Portugal to directly correlate driver motivation and alignment with a reduction in incidents and an increase in operational performance. In the same way, Claranet centralised the employee experience on the Planet platform, gaining reliable data to manage talent on a unified basis. By connecting gamification, evaluation and feedback, the platform delivers the insights needed to act before talent walks out the door. Request a demo to see these dashboards in action.

Further reading:

• People Analytics and Data-Driven Decisions
• How to Measure the ROI of Employee Engagement: The Practical Guide for HR Directors
• Skills Gap Analysis: Transforming Talent Management with Data and AI
• GFoundry TOAR: Manage Talent with AI, Data and LLMs
• The End of Annual Reviews: The Strategic Shift to Continuous Feedback in 2026
• GFoundry Case Studies