Business Insider: There is more talk than ever about implementing artificial intelligence in companies. From Lingaro's perspective, why is this technology so important today? How open are enterprises to it?
Tomasz Rostkowski:
We're definitely seeing growing interest in artificial intelligence, particularly since ChatGPT appeared. A growing number of companies understand that failing to invest in AI creates a real risk of being left behind. Most of our clients are global consumer goods companies that invest heavily in data analytics, and we at Lingaro have been helping them implement AI for many years. The organizations we work with are quite diverse — from large, technologically advanced enterprises to smaller businesses just beginning their AI journeys. Regardless of their size, the ones leading their markets are no longer asking "whether it's worth it" but rather "how to do it best." For them, AI implementation has stopped being a matter of choice — today it's a condition for maintaining competitiveness and effective growth.
Industry discussions increasingly involve descriptions of AI as traditional, generative, or agentic. How do you define these three categories, and what are the key differences between them from a business perspective?
Traditional AI means projects based on machine learning models that are built for client needs. These models use historical data to automate repetitive decisions in business processes. Typical applications include event prediction, forecasting future metric values, anomaly detection, or optimization. Implementations have clearly defined business goals expressed in expected improvements to efficiency metrics. Realizing these goals requires precisely defined operating contexts described by large, organized historical datasets. In these projects, the heavy lifting is mainly training the model and preparing the data. It's important to remember that investment in traditional AI is usually profitable when it concerns a large volume of repetitive processes relying on high-quality data. Since business scenarios where traditional AI brings benefits have been explored for years, organizations can usually refer to past experiences with similar implementations and know what results to expect.
In generative AI, the goal is to create new forms of content like text and images. Here we don't build a model from scratch on client data. Instead, we use large language models trained on enormous sets of data and add more or less advanced mechanisms to increase response accuracy. This approach greatly accelerates and simplifies the implementation of natural language processing solutions like chatbot assistants, tools for finding and synthesizing knowledge, or content generators. The large language models are ready-made, so these projects are mainly about integrating the models and building mechanisms that force them to stick to facts related to the use case scenario. For example, a model might be orchestrated to rely on knowledge embedded in company documents that were ingested into the model’s dataset. Generative AI is a relatively young and dynamically developing field. We have some proven scenarios where technology brings measurable benefits, but most projects tend to be considered investments in innovation.
Agentic AI is the next stage of generative AI development. Here we enhance large language models with additional logic and treat them as specialized agents that work together to perform more complex tasks. The distinguishing feature of the agentic revolution is the ability to build non-deterministic solutions where — unlike in traditional IT — we don't define and program the exact way to solve a task. The agents themselves handle these details; they emergently plan a sequence of actions and go on to distribute and execute the necessary activities. This approach is still young and has many challenges. Even so, it brings enormous automation potential — especially with respect to data analysis, employees’ daily tasks, and software development.
What are practical examples of agentic AI applications in companies?
Companies are mainly interested in agentic AI in these three areas I just mentioned.
First, in data analysis they aim to implement scenarios more advanced than are possible with the first stage of generative AI. With agents, it is possible to build Talk2Data solutions that let users ask more complex questions and get accurate, comprehensive answers — all in plain language with no technical expertise required.
The second emerging area is workforce automation. With respect to people’s day-to-day tasks, companies — and the end-users themselves — can create specialized agents that automate tasks in line with a self-service model. Ready-made agentic platforms are available from a range of technology providers. These platforms may not be mature yet, but they will be soon. Many companies are also already implementing their own architectures supporting agent development and maintenance, and we help them in doing so.
Third, AI-supported software development is the area where we see the most advanced tools available and the most real-world breakthroughs being made. It will still be some time before natural language is the only programming language you need to know, but at Lingaro we believe that new AI capabilities have a transformational impact on our industry. We're investing heavily in recognizing, exploring and adopting new approaches to delivery and operations.
What AI implementation strategies do you observe organizations following? Do their approaches differ depending on the type of AI?
Indeed, there are a wide range of AI implementation strategies, from “doubling down” with heavy investment in leading the race, to “wait and see” with a preference for proven practices and tools.
For traditional AI, there are plenty of industry-proven practices and tools to rely on. Companies leading the way are focused on scaling up, maximizing value, and minimizing costs. Companies earlier in their journeys can catch up quickly thanks to the established body of knowledge.
For generative AI, practices and tools are emerging rapidly. Discovering solutions with proven ROI remains a challenge, especially when there is pressure to do “something” to avoid missing out on a chance to build competitive advantage.
In this dynamic landscape, we can't be entirely sure what AI success will look like as tools, capabilities, and opportunities continue to evolve. What remains constant is that every AI solution depends on data as its foundation. That means companies should make it a priority to ensure that their data is ready for AI. They also focus on effectively managing streams of new ideas, prioritizing projects, building platforms that allow component reuse and control over the model ecosystem, implementing governance practices, and managing the adoption process.
A typical element of generative AI adoption strategy we observe is “AI democratization” by making AI tools available to a wide group of employees in a self-service model. When employees can experiment, they build new competencies and can independently discover useful applications.
At Lingaro, we observe that market leaders have been quick to implement their own controlled versions of generative tools to enable employees to learn and test new solutions from the bottom up in a safe environment. Simultaneously, from the top down they had expert teams analyzing potential use cases, evaluating them for business value, and prioritizing implementations. This winning combination of bottom-up and top-down initiatives allows for both rapid experimentation and strategic investment management. We advise our clients to seek this balance and have developed a framework for them to do so. We call it the GenAI Factory.
How do companies approach AI agent adoption?
If generative AI is the new frontier, then agentic AI is the next frontier where the phenomena we’ve discussed intensify. We have even fewer proven practices and tools here, but they are being developed at an unprecedented rate.
For generative AI, democratization is a way to learn and collect ideas. For agentic AI, on the other hand, democratization becomes one of the main drivers of change — mainly because agents encapsulate complex AI logic into an easily consumable entity that uses natural language to instruct and communicate. Here, the playing field shifts from applying AI to highly specialized decisions in high-volume processes (traditional AI) to automating massively diverse tasks and process elements of different job functions (agentic AI).
With respect to AI agents’ adoption, the current focus is on building self-service agent platforms, agent management platforms, and users’ competencies. In the future, the focus will shift to top-down initiatives to help navigate organization-wide process changes, operational models, and maybe even business models.
As I mentioned, data readiness is a critical factor for AI adoption. With agentic AI, this theme takes on added complexity. If we want to augment knowledge workers’ capabilities with automation, know-how regarding processes, standards, and best practices should be captured and digitally organized so AI can use it. Proper knowledge management practices around a company’s “DNA” will be essential here. Today’s agents can effectively perform activities like drafting documents, but the real art is creating agents that consistently perform at the level of the company's best experts in each field and maintain that quality even as business realities change.
How do you assess the future of AI in business? Will agentic AI become standard? Can we expect further breakthroughs?
Broadly speaking, the pace of civilizational development has been continually accelerating for a long time. The appearance of large language models leveled up this acceleration. We all struggle to take full advantage of the tools we now have available. Even so, we can assume that the next AI breakthrough will arrive faster than the previous one.
Whether agentic AI will become standard is hard to say; what it means has not yet been fully established. From a technical standpoint, agents are just a concept for organizing work with language models and, as such, may change.
On the other hand, I think the business concept of agents — as intelligent entities supporting peoples’ workflows and decisions — is here to stay. It will be forcing employees to evolve their competencies and maybe even devalue the ability to remember or analyze information. Organizations will transform into entities connecting human team members with AI agents.
Many people fear that agents will take over human jobs, causing unemployment among white- and blue-collar workers. Paradoxically, software development is the industry most impacted by the latest waves of AI tools. That's why I'm curiously observing how AI will affect the IT industry, which for the first time in history may find itself in a situation where labor supply exceeds demand. On the other hand, AI-powered software development will lead to greater demand for IT solutions as they become more accessible to the broader market. Will this balance out? Only time will tell.
Source: Business Insider https://businessinsider.com.pl/