Systems | Development | Analytics | API | Testing

This is a Data App that collects structured product submissions from a team, validates them, queues them for approval, and writes approved entries directly to a Keboola table. I built it with Kai in one conversation. No Google Sheets. No broken column headers. No emailing CSVs. If you've ever needed your team to submit structured data - new products, budget inputs, campaign briefs, vendor details - and the spreadsheet approach keeps falling apart, keep reading.

In the span of five days in March 2026, a single threat actor—TeamPCP—compromised a vulnerability scanner (Trivy), a code analysis platform (Checkmarx), and the most widely used LLM proxy in the Python ecosystem (LiteLLM). The attack chain was surgical: each compromised tool provided credentials to attack the next target.

In March 2026, security researcher isfinne discovered that LiteLLM version 1.82.8—the most popular open-source LLM proxy in the Python ecosystem, with approximately 97 million monthly downloads—contained credential-stealing malware published to PyPI. Within hours, version 1.82.7 was confirmed to carry a similar payload through a different injection method.

Last August, MIT released a landmark report that confirmed what many enterprise leaders had started to fear: most AI pilots are failing. After reviewing hundreds of AI initiatives, researchers found that 95% of generative AI pilots failed to reach production or deliver measurable results. The headline quickly hardened into a cliché: AI doesn’t scale.

Your developers are shipping more code than ever. GitHub Copilot, Cursor, and tools like them have fundamentally changed developer throughput - some teams are seeing 40-76% more code per person per sprint. That is the headline everyone celebrates. The part that keeps engineering leaders up at night is the other side of that equation: your testing pipeline has not changed at the same pace. Tests that used to gate two releases a week now need to gate ten.

Testing a FastAPI app that calls OpenAI, Anthropic, or Gemini gets expensive fast. The problem is not just the API bill in production. It is all the repeated traffic in development: prompt tweaks, CI runs, regression checks, and the load tests you keep putting off because every run burns tokens. Hand-written mocks do not help much once the app is doing multi-step LLM work.

It is Monday morning. Your VP of Engineering just forwarded a company-wide memo: every team needs to demonstrate AI adoption by end of quarter. At the same time, you learned last week that your QA budget was trimmed by 15%, because leadership assumes AI will "make testing more efficient." And your developers? Thanks to Copilot, Cursor, and Claude Code, they are now shipping 76% more code per person than they were two years ago.

IT leaders today are in a high-stakes gridlock between the pressure to invest in new AI solutions and legacy systems that aren’t equipped to support them. With the pace of work today, long modernization projects are often untenable. They disrupt your workflows. They use up your team’s resources. And they sometimes fail to deliver results. So how do successful organizations move forward? They rethink their approach.

Most teams know they are spending money on AI in production. Far fewer realize how much they are spending outside production. It’s easy to get lost as you evaluate which model has the best responses, is fast enough, and cheap enough to run in production. That is because the AI bill usually shows up as a giant blob. It is easy to see the total.

Your AI agent just placed 47 duplicate orders. It called the wrong API three times in a row. It looped through the same workflow for six minutes before anyone noticed. Nobody caught it in testing because nobody built the right tests. That's not a hypothetical. Enterprises using AI agents face this exact problem every week. The AI agent works perfectly in staging, but fails silently in production, and by the time the on-call engineer gets alerted, real customers are already affected.