I Used AI for 7 Days Here's What Happened

Day 1: Research Better Than Expected

I had a competitive analysis to produce for a D2C brand considering a new product category. Normally this takes me three to four hours: market sizing research, competitor mapping, pricing analysis, customer review mining. With Claude and Perplexity running in parallel, I had a first-pass analysis in 45 minutes. The market sizing figures were accurate I verified them against primary sources. The competitor mapping was comprehensive. The customer review synthesis from Amazon and Myntra was genuinely better than what I would have produced manually, because the AI could process 400 reviews in seconds while I would have read a representative sample of 30.What the AI could not do: tell me which of these competitors was actually winning in the market and why. That insight came from a 20-minute conversation with a category expert who had lived in this space for three years. The AI gave me the data. The human gave me the interpretation. The combination produced something better than either alone. Day 1 verdict: AI as a research tool is legitimately, significantly faster. The human judgment layer is still required and is still where the insight lives.

Day 2: Writing The Re-Prompting Problem

I needed to write a detailed product brief for a new feature. I spent 40 minutes trying to get an AI-generated brief that matched what I actually needed. The first draft was generic. The second was better but missed the technical constraints. The third was technically accurate but wrote in a tone that would have confused the engineering team it was aimed at. By the time I had a draft I was willing to use, I had spent more time than it would have taken me to write the brief myself. The output was fine. The process was inefficient.The lesson I drew from Day 2: AI writing assistance works best when I invest ten minutes in writing a detailed, specific brief before I prompt one that captures the audience, the tone, the specific constraints, and the outcome I want the document to produce. When I did this on Day 5, the first draft was 85% usable and took me 15 minutes to edit into a final version. The re-prompting problem is mostly a briefing problem. AI cannot read your mind. The more precisely you describe what you want, the less time you spend asking for it again.

Day 3: Email The Embarrassing Incident

I used Copilot to draft a follow-up email to a founder I had met at an event. The AI produced a professional, well-structured email. It also addressed the founder by the wrong name it had pulled a different contact from context and referenced a detail about their company that was incorrect based on the cached information in my CRM. I caught both errors in review and corrected them before sending. But the experience made me viscerally aware of something I had understood abstractly: AI email assistance is producing confident, well-formatted outputs that may be factually wrong about the specific person you are writing to.Day 3 verdict: for bulk email drafting where the recipient details are consistent and verifiable, AI is genuinely useful. For relationship-sensitive, high-stakes communications where specific person details matter, the verification step is non-negotiable and cannot be rushed. The 30 minutes per week that Microsoft reports Copilot saving on email is real. The 10 minutes per week I spent checking AI-generated details about specific recipients is not in that figure.

Day 4: Planning Genuinely Useful

Day 4 was a quarterly planning session where I needed to structure a 90-day roadmap for a client's operations team. I gave Claude the context team size, current tools, top three operational problems, the outcomes we wanted to achieve and asked it to produce a structured 90-day deployment plan. The output was the best first draft of a consulting deliverable I had ever received from any tool. It was structured logically, the phase sequencing made sense, and the milestone definitions were specific enough to be actionable.I spent 30 minutes editing it adjusting phase timelines based on my knowledge of this specific client's change management capacity, adding two constraints that the AI did not know about, and sharpening the success metrics. The final document took me one hour total. Without AI, this deliverable takes me three to four hours. Day 4 was the day I genuinely understood what AI leverage feels like when the context is rich and the task structure is clear.

Day 5–7: The Patterns That Emerged

By days 5, 6, and 7, I had developed a working model of when AI makes me faster and when it slows me down. AI makes me genuinely faster on tasks that are well-defined, where the output format is standard, and where I have enough domain knowledge to evaluate the output quickly. Research synthesis, structured document drafting, data analysis framing, agenda preparation, and summarisation all fall into this category. I conservatively estimate I saved between 8 and 12 hours across the week on these tasks.AI slowed me down on tasks requiring contextual judgment about specific relationships, tasks where the quality bar required multiple revision cycles because the brief was inherently ambiguous, and tasks where the verification burden was high enough that checking the AI's work took longer than doing the work directly. The honest net time saving across the full week was approximately 6 to 8 hours not the 20 hours productivity gurus claim, but real and meaningful.

What I Changed Permanently After the 7 Days

• I now write a detailed prompt brief before generating any AI output longer than a paragraph audience, tone, specific constraints, desired outcome. This investment in upfront clarity eliminates most of the re-prompting cycle.
• I treat AI research outputs as a starting point requiring primary source verification for any claim I am going to use in a client-facing context the AI is a fast first pass, not a final source.
• For high-stakes communications to specific individuals, I review AI drafts with the same scrutiny I would apply to a junior team member's draft the factual details about the specific person and their company require human verification every time.
• I use AI most aggressively for the tasks that were previously most tedious: first drafts of structured documents, research synthesis, and agenda preparation. These are where the leverage is real and the verification burden is low.
• I stopped expecting AI to have the contextual judgment I have accumulated over years. I now use AI for the mechanical and structural parts of my work and invest my own judgment in the parts that require it.