Garage Eight: how AI is changing the work of analysts and why junior positions are disappearing

Garage Eight analyst Vladimir Syropiatov described six AI trends that are already changing everyday data work. His main thesis sounds harsh: neural networks don't eliminate the profession, but quickly devalue routine and raise the bar for entry.

Who AI will replace

According to the author, junior analysts feel the strongest pressure from AI. Work that was recently considered normal entry-level tasks—ad hoc requests, simple dashboards, draft reports, basic SQL and Python debugging—is now increasingly done in minutes with the help of a model. Companies continue to hire analysts, but increasingly search less for people to handle routine tasks and more for specialists who can provide the model with the right context, verify the result, and connect it to business decisions. The author proposes a simple formula: one mid-level analyst with AI replaces several juniors.

• typical junior positions tied to routine are disappearing;
• managers themselves cover part of analytical questions through AI;
• vague formulations give way to clear prompts and specifications;
• training shifts from syntax to logic, validation, and LLM work;
• demand grows for hybrid roles at the intersection of analytics, domain expertise, and AI.

The author sees a separate effect in value redistribution. Cheap AI tools give small and medium-sized businesses access to capabilities that were previously only available to large teams with dedicated analytical resources. In this scenario, an analyst is needed not as someone who simply "pulls numbers," but as a solution architect: explaining data limitations, gathering context, verifying the model's output, and ensuring that a quick machine answer doesn't turn into an expensive business mistake.

Garage Eight in practice

To show this isn't just theory, Garage Eight describes several internal cases. For working with sensitive data, the team tests local models, including DeepSeek-R1 14B, so managers and analysts can get SQL suggestions and analyze conversion drop causes without risking data leaks. According to the author, in this scenario, the speed of gaining insights increased 32 times, and employees better understood the origin of metrics themselves because they started working with them directly rather than only through queues of requests to analysts.

Three more applications look equally pragmatic: AI extracts meeting summaries, validates code before deployment, and helps compile draft reports and presentations. The article claims that AI validation speeds up error detection 3–5 times and reduces production bugs by 70%, while report preparation drops from approximately three hours to one. Inside the company, this has already become a set of best practices: run code through a local LLM and review presentations together with AI before showing them to the audience.

"AI won't replace analysts.

But analysts who use AI will replace those who don't."

How to prepare now

From these observations, the author draws two practical conclusions. For managers, the advice is simple: don't wait for the perfect moment, but break down processes into parts and implement AI step by step. In parallel, you need to train your team, negotiate with security and compliance, and create an environment where you can quickly test a hypothesis rather than spend weeks aligning on a pilot. At Garage Eight, they're already looking at automatic task prioritization, personal AI assistants, AI mentors for newcomers, and regular searches for non-obvious hypotheses in data.

For analysts themselves, the plan is also clear. Learning tools needs to start now, but not at the level of "press button—get answer," but at the level of critical verification, product thinking, and communication. The more routine a model takes on, the more valuable skills become—explaining complex things in simple words, seeing the process behind numbers, and understanding where AI can be trusted and where its output needs hard rechecking. Otherwise, the specialist risks competing not with a colleague, but with a cheap and very fast automated layer.

What this means

The Garage Eight article captures well the shift already visible in many data teams: AI doesn't cancel analytics but moves it up the value chain. Winners will be specialists who can combine model speed with their own understanding of business, data, and risks. Losers will be those who continue to sell the market only manual routine that neural networks have already learned to do faster, cheaper, and with almost no queue for execution.