I recently helped our software engineering team rethink our development workflow. With AI now being used heavily both for prototyping solutions and for development itself, we realized we needed to revisit our Software Development Life Cycle.
The first step was to do deep research on the topic, to understand the best practices from people who are actually doing this. Instead of researching and putting together a report by hand, I decided to use Claude Cowork — and the result was phenomenal. In a single session, I produced a detailed report, with sources, ready to share with the team.
Continuing with my goal of helping you automate manual tasks — like I did in the posts about automating meeting notes and automating presentations — in this post I’ll walk you through the workflow I’ve been using to do any kind of deep research with AI and turn it into a presentation-ready report.
What you need to follow along
To produce reports from deep research, you’ll need a paid Claude subscription. The reason is simple: even if you’re on the Free plan, the research process burns through a lot of tokens — Claude will browse many websites, synthesize information and produce long documents, which easily hits the limits of the free version.
My recommendation is to subscribe to the Claude Pro plan, at $20/month. It’s the same plan you’re already using if you followed the methods in my previous posts.
If you don’t have the app installed yet, download it from the official website. Every feature we’re going to use is only available in the app — not in the browser.
What Claude Cowork is
Claude Cowork is a Claude Desktop feature that lets you delegate complex tasks for the AI to work on autonomously. You describe what you need, Claude builds an execution plan and works in the background — searching the web, analyzing sources and producing documents — while you do other things.
In practice, Cowork turns Claude into something like a junior analyst. You give it the brief, it runs the research and hands you a structured report in Markdown. The difference is that it all happens in minutes, not hours.
Here’s how the workflow goes:
- You describe the topic — it can be a research question, a problem to investigate or a topic to map out.
- Claude builds a research plan — it shows you the topics it plans to investigate and the sources it will check.
- You approve or adjust the plan — this is the most important part. Before it starts researching, you can redirect it, ask it to include a specific angle or remove something that doesn’t make sense.
- Claude runs the research — it goes through dozens of sources, pulls out the relevant information and synthesizes everything.
- You get the report — a Markdown document (
.md), structured with sections, references and conclusions.
How to use it in practice
Step 1 — Open Cowork and describe the topic
In Claude Desktop, start a new conversation and turn on Cowork mode. Describe your research topic clearly, including the context and what you expect as the outcome.
Example of an initial prompt
“I need to do deep research on how tech companies are adapting their software development processes (SDLC) to incorporate generative AI. I want to understand: which stages of the SDLC are most affected, which tools are being adopted, and what the emerging best practices are. The goal is to present the conclusions to my engineering team.”
The more context you give, the better the research plan will be. Include details like who the report is for, how deep it should go and whether there’s a specific angle you want to explore.
Step 2 — Review the research plan
Before running anything, Claude will show you a structured plan with the topics it intends to research. This is the moment to adjust.
Don’t skip reviewing the plan. This is where you make sure the research is heading in the right direction. Asking to add or remove a topic at this stage takes seconds — redoing an entire research run takes minutes.
Review the proposed topics and, if needed, ask for changes:
- “Add a topic on the risks and limitations of using AI in the SDLC”
- “Remove the part about no-code tools, it’s not relevant to my context”
- “Go deeper on AI-assisted code review, that’s what my team cares about most”
Step 3 — Let Claude do the work
Once you’ve approved the plan, confirm and let Claude run it. It will research autonomously across dozens of sources, pull out the relevant information and start putting the report together.
This step takes a few minutes, depending on how complex the topic is. Meanwhile, you can keep doing other things — Claude works in the background and notifies you when it’s done.
Step 4 — Review the generated report
When it’s finished, Claude delivers a Markdown document (.md) with all the research synthesized. The report usually includes:
- Executive summary with the main conclusions
- Thematic sections organized by topic
- Sources and references cited throughout the text
- Recommendations or next steps, when applicable
You can open the .md file directly in your browser, in Obsidian or in any text editor. If you use Obsidian like I do, the file is already integrated into your notes and searchable.
If you don’t use Obsidian, no problem. .md files can be opened in VS Code, in the browser, or converted to PDF with online tools. The format is lightweight and compatible with pretty much everything.
Why Claude Cowork and not Perplexity?
If you already use Perplexity for research, you might be wondering: why switch? The short answer is that they’re tools with different purposes.
Perplexity is excellent for one-off questions — when you want a quick answer with sources. But for deep research that turns into a report, Claude Cowork has significant advantages:
Control over the process vs. an instant answer. Perplexity gives you a ready-made answer. Claude Cowork lets you see and steer the research plan — you can iterate, ask it to dig deeper into a specific angle, or redirect it before it starts writing.
Output as a reusable artifact. The report Claude Cowork generates is a complete .md file that can become the input for other processes — a presentation, a post, a briefing, or even a prompt for another agent. With Perplexity, you get an answer you have to copy and paste.
Integration with your workflow. Claude Cowork lives in the Claude app, installed on your machine, right next to your files. The research can be combined with automations — for example, saving it to Obsidian automatically, generating a slide deck right after, or feeding another document. Perplexity is an isolated tool.
Transparent reasoning. Claude Cowork tends to spell out what it researched, why it chose a given angle and what it found in each source — which is useful when rigor matters.
Iterative depth. With Claude Cowork you can do multiple rounds in the same session — initial research, spot the gaps, dig into specific points. Every question in Perplexity is essentially a fresh start.
I’m not saying Perplexity is bad. I use it myself for quick day-to-day research. But for research that’s going to become presentable material, Claude Cowork is much more efficient.
Tips for better results
After using this workflow for a while, I’ve learned a few things that make a big difference in the quality of the output.
Ask for a first version and iterate. Don’t try to get the perfect research on the first try. Ask Claude to generate a first version of the report. Read it, spot the gaps, and ask it to dig deeper into specific points and produce a version 2. Every round of iteration makes the material richer.
Example of a prompt to iterate
“The report turned out well, but I need you to go deeper on the section about AI-assisted code review. Research specifically how companies are using LLMs to review pull requests, which tools exist and what results have been reported. Generate a version 2 of the report with that section expanded.”
Give context about the audience. Saying “the report is for a presentation to directors” produces a different output than “it’s to share with the engineering team”. Claude adjusts the level of technical detail and the language based on that.
Use the report as a starting point, not the final product. Just like in my previous posts, the AI gives you an excellent starting point. Review it, adjust the tone to your reality, and add your own conclusions. The value is in saving hours of research and structuring, not in replacing your judgment.
Keep your reports organized. If you use Obsidian, create a dedicated folder for research. Over time, you build up a searchable library of reports that serves as a reference for future decisions.
Next steps
Once the .md report is ready, the natural next step is to turn that content into a presentation. Claude can help you condense the key insights into slides and, using a brand identity Skill, generate a presentation that’s ready to go in your company’s branding — as I explained in the post about automating presentations.
The full workflow looks like this: deep research → Markdown report → PowerPoint presentation. What used to take a whole day now takes less than an hour.
I plan to explore other variations of this workflow in future posts — like using the same method for competitive analysis, product discovery and technical investigations.
If you try the method, let me know how it goes.