What’s on Our Radar 

For a while, the central question driving discussions at GX2 was whether companies were integrating AI into software development or still operating as they did in 2020. That conversation has evolved, and the question now is different and more uncomfortable: Why are so many companies still using AI as a chat assistant when it’s already capable of doing the work?

In recent weeks, product launches and industry news have made it clear that the market has moved to a new level. In a short period, we’ve seen personal AI agents reach consumers, WhatsApp Business open up integrations with AI agents, and drones begin making commercial deliveries in Brazil. At GX2, we’ve been following these developments closely, and one thing has become clear: the differentiator is no longer whether you use AI, but how deeply it’s integrated into your business.

The Current Landscape: From Conversation to Execution

Conversational AI is passive by nature. The user asks a question, waits for an answer, and asks again. The user remains part of the execution process: they’re the ones sending the email, updating the CRM, and entering information into the ERP.

Agentic workflows change that dynamic. The user defines the goal, and an agent, or a group of agents, works through the steps needed to achieve it. Humans remain responsible for making decisions, while agents handle the execution.

Recent launches illustrate this shift:

• GPT-6 Astra (OpenAI): In a demonstration shared by the company, tasks such as scheduling an appointment or searching for an apartment, tasks that could take a person one to six hours, were completed in minutes. When a decision is required, the agent pauses and hands control back to the user.

• Meta Muse: A personal agent integrated into Meta’s ecosystem, capable of interacting with emails, calendars, and messages on the user’s behalf.

• WhatsApp Business + MCP: AI agents can now interact directly with the platform, eliminating many of the manual steps traditionally involved in approval and integration processes. What once took hours can now take minutes.

MCP is the protocol that allows agents to access tools and data. When tools become accessible to agents, not just people, the game changes.

What This Means for Business

3.1 From Software as a Service to Service as Software

Sequoia Capital, an early investor in companies such as Apple, Google, and WhatsApp, and more recently OpenAI and NVIDIA, has published an investment thesis that captures this shift: Service as Software.

Instead of selling customers a tool they use to achieve a goal, the emerging model is to sell the end result, with software serving as the means of delivering the entire service from end to end.

The same material shows that approximately 49.7% of AI usage today is concentrated in software engineering. Other areas, including back-office automation, remain relatively underexplored, and that’s where the opportunity lies.

3.2 Connected Data, With Governance Built In

RAG (Retrieval-Augmented Generation) already enabled AI to retrieve documents and generate answers based on company information. However, each query was typically limited to isolated sources, such as a contract or a user manual.

Graph RAG takes this approach a step further. Instead of working with disconnected documents, AI can access a network that connects emails, CRMs, ERPs, and cloud drives, correlating data from different sources.

Governance is what makes this next step possible. AI operates within each user’s access permissions: an HR employee can access payroll records, while a systems analyst cannot. Without these controls, connecting everything would introduce significant risk. With them, data stops being a collection of isolated silos and becomes an asset AI can use securely.

3.3 Smaller Models, Closer to Where Decisions Happen

Not every problem requires a massive model running in a data center. Small Language Models (SLMs) can run directly on hardware, delivering low-latency responses, something essential for autonomous vehicles, hospitals, and drones. In these cases, before choosing a solution, the right question to ask is: “Which model is best suited to this problem?”

3.4 A New Market Trend: The AI Roll-Up

Investors are acquiring small and midsize companies that are financially healthy and operate effectively without AI, with the goal of consolidating them and introducing layers of agentic automation. The message to the market is clear: AI maturity has become a driver of both business value and competitiveness.

From Chat to Agent

The question used to be whether we were using AI. Now, it’s whether we’re ready to work with agents. Moving from tools to agents requires three things at once: methodology (installing an assistant isn’t enough), governance (keeping data and permissions under control), and a focus on business value (solving real problems rather than chasing hype).

This is the thinking behind the GX2 AI Platform, which takes software development from Coding Assistant to Agentic Coding. Instead of relying on an assistant to support developers who are already writing code, specialized agents work across the entire development lifecycle, from discovery to deployment. Each agent has its own scope and quality criteria, shares the customer’s business context, and operates under human validation.

Built on this foundation is Augmented Squads: a hybrid, multidisciplinary team in which human engineers and autonomous AI agents work together as a single system. And the benefits go beyond promises. Performance is contractually guaranteed, with at least 80% higher productivity compared with a traditional squad operating without AI.

Is your team ready to work with AI agents? If you’d like to discuss this with GX2, leave your contact information, and we’ll get back to you. Prefer a more direct approach? Talk to our AI agent. It responds in real time, with no form to fill out and no waiting.