Pixelwolf reinvents its operations with AI after a decade of evolution

From software house to AI consulting and implementation

Founded in 2015, Pixelwolf began as a software development company, creating apps and websites. At that time, the Brazilian market wasn’t yet discussing user experience in a structured way, so the company started offering a free discovery process to understand client needs before writing code.

This approach caught the attention of clients like Jovem Nerd, XP, EBANX, Hotmart, and Porta dos Fundos. By 2018, the company was already working on AI projects alongside giants like IBM and Microsoft, before the topic gained its current prominence.

One of the company’s custom products, Wollo, was acquired by Hotmart in 2020 and integrated into Sparkle, an app launched in September 2019 for distributing and selling online content, courses, and events. In 2021, following the acquisition, Pixelwolf became the embedded technology team for Fluency Academy, a language school backed by General Atlantic.

The integration lasted until the end of 2024, when Fluency stopped using the brand. Thiago Bisewski, the company’s founder, relaunched Pixelwolf and re-entered the market amid the rise of generative AI.

Why most AI projects fail

A study by MIT’s NANDA research group helps explain the situation Bisewski describes. According to the research, nearly all organizations analyzed—95%—failed to demonstrate measurable financial returns on their investments in generative AI.

Among successful initiatives, those carried out in partnership with specialized suppliers were twice as likely to succeed compared to those developed entirely in-house.

For Bisewski, the problem usually starts before the technology. “I see many companies thinking they need to use AI and immediately start implementing tools without knowing which problem, cost, or indicator they want to change,” he says, describing the corporate fear of falling behind.

The founder of Pixelwolf lists four common mistakes:

1. Adopting AI without a defined problem. The company decides it “needs to use AI” and starts implementing tools without knowing which process, cost, or indicator it wants to modify. This is the main reason many projects fail to generate returns.

2. Releasing tools without governance. Without usage and security policies, the result is data leaks and inaccurate information in reports.

3. Blocking everything after the first incident. This reaction leads teams to turn to “shadow AI”: personal tools that escape the company’s control.

4. Building everything in-house. The same MIT study shows that projects with specialized suppliers had twice the success rate compared to those developed solely by internal teams. Anyone who has implemented the technology multiple times knows what can go wrong.

How process diagnosis works

Before any implementation, Pixelwolf conducts a process-focused consultation. The immersion lasts four to six weeks and includes interviews with directors, managers, leaders, and operational teams, without discussing technological tools at this stage.

The diagnosis is measured by performance indicators and return on investment. For example, a manual process that consumes weekly team hours can be converted into an annual cost estimate, helping to prioritize where AI will have the greatest impact.

From there, the company creates a prioritization matrix balancing quick wins and long-term projects, with deliverables organized in twelve-week cycles. Pixelwolf handles the implementation of selected agents and automations, while the client monitors the indicators defined in the diagnosis.

A small team supported by AI agents

Pixelwolf currently operates with twelve senior professionals, each managing between five and six AI agents simultaneously. According to Bisewski, this change has transformed the routine of programmers.

“Before, code development was almost therapeutic. Today, it’s a job of review and architecture,” says the founder, who also teaches in the MBA program in AI at Saint Paul Business School.

Pixelwolf’s journey, from a software house to AI consulting and implementation, reflects a broader trend in the tech market: AI adoption increasingly depends less on specific tools and more on how companies organize their data and internal processes before automating any step.