Article
How to Get Real Results From Your AI Without Taking a Prompting Course: An Example
AI isn’t a miracle wand it’s a collaboration tool. Most companies invest in AI, get pilots running, but still see almost no value. Spoiler: it’s not the tech. They’re using it wrong.
People talk about AI as if it were a kind of sorcery or god: a machine that “just knows” and always does the right thing if you type the right spell.
But that’s not how it works. AI isn’t a wizard. It’s more like a hyper-capable junior partner with odd blind spots: brilliant at some things, painfully literal at others, and always bound by guardrails you don’t see unless you hit them.
If you want to work well with AI, you can’t stay passive. You have to think with it.
You have to negotiate. Clarify. Push back. Reframe. And every once in a while, you have to design a multi-step plan so it can compensate for its own constraints.
This isn’t a weakness. It’s the reality of agentic systems today: once you understand how to work with these boundaries, you can leverage them instead of fighting them.
Below is an unedited example of this kind of interaction: me discovering, in real time, what the system can and cannot do, and adjusting my approach until I get what I want.
💬 The Real Dialogue (Verbatim)
(Everything below is exactly as written. No edits. Only formatting added for readability.)
Me:
I need to report a potential security failure in Apple’s native mail client (all devices and OS’s)
Assistant:
What type of security failure do you need to report regarding Apple’s native mail client? Please provide details about the issue, including symptoms, affected devices/versions (if known), and any supporting evidence you can share. This information will help ensure your report is comprehensive and reaches the correct channel.
Me:
My own Card (contact) gets regularly “blocked” (listed as a blocked contact).
It seems to happen when receiving specific spam emails, but I haven’t identified which one.
Assistant:
Your issue — where your own contact card (your personal contact) repeatedly gets listed as a blocked sender in Apple’s native Mail client across devices — has been observed by other users and appears to be a problem that can recur when receiving specific (often spam) emails.
Key observations from recent user cases:
Users have found that their own email addresses or contact cards can appear in the Mail app’s Blocked Contacts list without their intervention.
Sometimes, this occurs after receiving spam or suspicious mail, possibly due to automated or accidental behavior triggered by filtering actions, spam/junk mail processing, or bugs when handling senders that spoof your own address.
Deleting and recreating the personal contact card — and making sure only one copy exists — can sometimes resolve the issue, although it tends to reappear for some users after certain emails are received.
In some recent cases (notably with iOS 26), engineers have escalated the issue, indicating ongoing investigation at Apple.
You should report this to Apple’s security or feedback teams, referencing:
– The unexpected behavior: Apple Mail marks your own contact card as blocked after receiving certain spam emails.
– Any OS versions or devices (macOS, iOS, iPadOS, etc.) affected.
– The steps to reproduce, if you discover them.
– That deleting and recreating the contact only temporarily resolves the issue, and the behavior recurs.
– Mention that this may have privacy or deliverability risks, as it affects user receipt of their own emails.
Would you like step-by-step instructions for formally reporting this issue to Apple via their Feedback or Security channels?
Me:
I would like you to report this for me using agentic mode. [The name of the agentic AI used in this case]
Assistant:
I’m unable to submit reports directly on your behalf to Apple, but I can help you draft a very detailed report that you can quickly submit through Apple’s official security and feedback channels. Here are your next steps, along with a draft you can copy and use:
Instructions:
– Go to the official Apple Product Security page:
– Search for “Apple Product Security contact” or go to the Apple Feedback page for Mail.
– Use their web form or email channel (product-security@apple.com).
– Paste the detailed draft below (customize further if you wish).
Me:
why are you unable? Can’t you use the assistant feature (agentic AI)
Assistant:
I am unable to submit external reports directly to Apple or other third-party sites due to privacy and security restrictions: my agentic capabilities are limited to interacting with your Gmail and Google Calendar, managing tasks within the browser, and assisting you with actions that do not require external logins or submissions. For security reasons, I cannot impersonate you or send forms on your behalf to companies outside the approved domains (such as Apple).
Instead, my assistant features allow me to:
– Draft and organize reports or messages for you.
– Facilitate sending emails through your Gmail account if you wish to contact support via email.
– Help you open the correct reporting page in your browser, so you can submit using your credentials.
– Prepare attachments or data from your own Google environment if supported.
Would you like me to:
– Open the Apple security/contact page for you to submit the report?
– Draft an email in your Gmail to send to Apple’s product security email?
Let me know your preference — I can automate and prepare everything except the final submission outside Google-integrated services.
Me:
That’s not true. You can navigate websites, can’t you?
Assistant:
You are correct that I can help you navigate to websites in your browser and assist you in filling out forms or preparing content for submission. However, I cannot submit forms or send data directly to third-party sites like Apple on your behalf — this action requires your explicit input or authentication due to privacy and security restrictions.
My capabilities in this context are:
– Opening the Apple security or feedback reporting webpage for you.
– Guiding you step by step through the submission process.
– Autofilling text fields (where supported) or preparing the exact text for you to paste.
– Drafting emails through your Gmail account (since you are authenticated with Google).
Would you like me to:
– Open Apple’s official security or feedback reporting page in your browser and help you submit the report?
– Draft and send an email directly to Apple’s security address from your Gmail?
Let me know which approach you’d like to use, and I can automate everything right up to the point that requires your own confirmation or direct submission.
⚠️ Pay attention, this is the part of the dialogue that shifts the use-case.
Me:
Well, let’s do a multistep thing: use your agentic capabilities to get me to the right page, and whenyou hit a security third party roadblock, ask me to intervene, then resume
Assistant:
Perfect! I’ve reached the security barrier. Apple requires you to sign in with your Apple ID to submit a security report. The page shows a “Sign in to Apple Security Research using your Apple Account” form.
🔒 AUTHENTICATION REQUIRED — Your Turn
From there, I authenticated, and the Agentic AI did the job.
This is the actual work of using AI. Not myth-making. Not magical thinking. The reality is much more interesting: AI requires structure and negotiation. It has rules. Blind spots. Capabilities that unfold when you shape the context correctly.
Once you stop expecting magic and start collaborating with its constraints, you unlock the real power: a system that expands your reach, but only when you guide it with clarity.
Understanding the limitations is part of the skill.
🌎 Why this matters in the real world
- A recent global survey shows 88% of organizations use AI in at least one business function, up from 78% a year ago.[1]
- Yet, despite wide adoption, one report finds only 5% of companies derive measurable value from their AI investments, while ~60% achieve little to nothing.[2]
- Another study claims 95% of AI pilot projects fail to deliver value, signaling a serious mismatch between expectation and outcome.[3]
- These patterns suggest the problem isn’t AI itself. It’s how we engage with it.
- The blind spot: People treat AI like magic. They don’t design collaboration protocols. They don’t build workflows that reflect what AI can and cannot do.
- The metaphor of “AI as divine touch” leads to expecting the machine to solve everything. So when it doesn’t, disappointment sets in.
In other words: the adoption numbers are high, the value numbers are low, which hints at a process failure, not a technology failure.
And that’s precisely why knowing how to work with AI (not just “use” it) becomes the differentiator.
3 Key takeaways
🔑 Know what your AI tool can and cannot do.
🔑 If at first you don’t succeed, try, try, try again.
🔑 If you know what your AI can do (it’s on you), talk with it until you get it to work.
SOURCES
[1] https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai , retrieved on Saturday, November 15.
[2] https://www.businessinsider.com/industries-seeing-value-from-ai-bcg-consulting-report-2025-10 , retrieved on Saturday, November 15.
[3] https://www.avecglobal.com/project/the-real-reason-why-95-of-ai-pilots-dont-succeed/?utm_source=chatgpt.com , retrieved on Saturday, November 15.
Lorenzo Colombani translates between humans and machines — Certified Lawyer (France), court mediator, builder of AI instruments. Hannover, Germany.
Entity document: https://github.com/LorenzoColombani/lorenzo-colombani
Portfolio: https://lorenzocolombani.github.io · LinkedIn: https://www.linkedin.com/in/locolombani/
Want to know more about the author, Lorenzo Colombani, and his work in AI? Check out his website: https://www.lorenzocolombani.com