Howdy wizards,
In this edition:
OpenAI makes it possible to put Deep Research into your app
What context engineering is, and how it’s different from prompt engineering
Pour yourself some quality bean juice and find a comfy spot.
Here’s what’s brewing in AI.
DARIO’S PICKS
Via: OpenAI for Business on LinkedIn
OpenAI came full circle this week by bringing it’s Deep Research feature to the API. This means devs and businesses can now build deep, multi-step analysis right into existing tools and workflows.
A couple of weeks back, OpenAI opened ChatGPT to connectors from third-parties, including companies like Microsoft, Google, HubSpot, and GitHub—which makes it possible to use Deep Research on external data. Now they're giving companies the reverse option - integrating Deep Research directly into their own apps.
Companies are currently using Deep Research for everything from legal analysis, to economic research, and private equity due diligence — complex tasks that typically require a ton of manual work.
Deep Research uses a variety of tools to give you maximum context and depth on research and analysis tasks. It can use web search, is capable of accurately analysing data (code interpreter) and presenting information in specific ways such as nice tables (structured output). It also runs in the background so you can keep chatting and it’ll ping you when it’s done (takes 10-15 min usually).
If you’re curious about embedding Deep Research in your app, check OpenAI’s full cookbook here.
Why it matters Companies have lots of internal tools and purpose-built UIs, for which it may be more practical to bring Deep Research into those tools rather than creating an MCP connector and bringing their data into ChatGPT.
For companies with privacy concerns, using the Deep Research API (vs merely sending data to ChatGPT) could provide materially finer-grained privacy control, such as being more deliberate about which parts of the data it has access to, compliance logging and avoiding stuff getting saved into ChatGPT’s memory.
Also worth keeping in mind: DR can produce impressive-looking but factually incorrect data. It’s key to isolate and articulate your problem carefully. Or you might spend more time fact-checking than just doing the research yourself.
IN PARTNERSHIP WITH TAPLIO
In the sea of tools to help you with LinkedIn growth, Taplio is one of the few that actually moves the needle. Their powerful new features help you create faster, engage smarter, and grow easier.
🔹 Likers & commenters dynamic lists
Turn post engagement into warm leads. Instantly build contact lists from people interacting with your content—and export them to your CRM or lemlist in one click.
🔹 Add prospects from LinkedIn search
See a list of ideal profiles? Select and add them straight to your Taplio contact lists—right from LinkedIn.
🔹 Engage with saved posts
Save your best posts and engage with them all in one place. Less time, more meaningful conversations.
Some features are exclusive to certain plans—so you’ll want to check what’s available at each tier. If you’re serious about LinkedIn growth, now is the perfect time to test it.
UP CLOSE
This section is about how I’m using AI from week to week, as well as practical tips & tricks I discover and actually use.
TLDR: Handing over your entire database to AI can be amazing for exploratory analysis. But for solving specific problems, you’ll probably want to make deliberate choices about what goes into the context.
The term context engineering is trending. Simply put, it’s giving the AI all the information it needs to help you effectively. This is not an easy task because amount, relevance, modality and timing of the information matters, and it requires your judgement on what to include and what to leave out.
Providing the right context is essential for creating apps and repeatable workflows using AI, but also highly relevant in everyday ChatGPT/Claude/Gemini usage.
I think it’s a more useful approach than traditional prompt engineering for a few reasons:
LLMs are getting better at understanding your intent, so the exact instructions you give it become less relevant (although they’re still important).
Understanding how much information to give an LLM is important and non-trivial. If you overload an LLM with too much context it doesn’t know what to prioritise and you’ll get a crappy output. Give it too little and you’ll get a lot of thin assumptions baked into its response, which might not be what you want.
Relevancy of the information is key. If it’s outdated, low-quality, formatted in a way that’s not desirable, etc. the results will be accordingly. Considering this in your approach includes providing few-shot examples that show the AI exactly what good output looks like for your specific situation, rather than letting it guess.
Context engineering encourages breaking down your problem into steps and clearly defining your context requirements at each step.
EXAMPLE: prompt engineering vs context engineering
Here's an example showing the difference between focusing on instruction design (prompt engineering) vs information environment design (context engineering) for a specific task.
Prompt engineering (ie crafting precise instructions to get better outputs):
Design instructions | Throw in a bunch of context |
|---|---|
"You are a strategic business analyst. Analyze our Q3 performance. Focus on actionable insights. Provide recommendations with for Q4 improvement." | 📎 Q3_financials.xlsx, competitor_report.pdf, everything else that seems relevant |
This approach focuses on crafting good instructions and providing data for context. Advanced prompt engineering could also break this into steps with different instructions at each phase. But the core approach is still about optimizing the instructions rather than strategically designing the information environment.
Context engineering (designing what information AI has at each decision point):
Break up problems into steps | Curate context based on the next step |
|---|---|
STEP 1 "Analyze our Q3 performance" | 📎 Q3_financials.xlsx + "We're B2B SaaS, 150 employees, missed revenue target by 12%" |
STEP 2 "What's driving our performance issues?" | 📎 Customer churn data + support tickets + "Churn up 3% but pipeline strong - need to understand disconnect" |
STEP 3 "What are our improvement options?" | 📎 Budget data + team capacity + strategic priorities + "Q4 budget frozen, no new hires, but can reallocate resources" |
STEP 4 "Given all this context, what are our top 3 actionable Q4 moves?" | 📎 3 examples of previous quarterly action plans + analysis from previous prompts retained |
Instead of generic business advice, this might get you specific recommendations like “Move 2 engineers from Feature X to fix the onboarding issue causing 40% of the churn”.
You could use more elaborate prompts at each step too, but the key insight is about curating and deliberately controlling the context rather than just improving the instructions.
“in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step”
Becoming good with AI doesn't stop at prompt or context engineering, though. These involve understanding LLMs, articulating a task accurately and making strategic choices about how the supporting knowledge (context) gets packaged for AI consumption. Important—but only as good as your underlying understanding of the problem and clarity of thinking.
FEATURED GPT
The Business Launcher GPT by Wix helps users generate personalized business ideas, company names, and business plans based on their skills, interests, and goals for starting new ventures or side hustles.
Want your custom GPT featured? Apply here.
THAT’S ALL FOR THIS WEEK
You've somehow made it all the way to the footer. Either you're genuinely interested or you're the type who reads shampoo bottles. I’m curious — what do you like about my newsletter? What can I do better?
Hit reply and lmk.
Was this email forwarded to you? Sign up here.
Want to get in front of 17,000 AI enthusiasts? Work with me.
This newsletter is written & curated by Dario Chincha.
Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.