"AI agents" are a shibboleth for "no-code"

shibboleth: a custom, principle, or belief distinguishing a particular class or group of people

I was attending an AI Agent conference to see what the (mostly self-professed) experts in the field are working on, and suddenly I got a ping of enlightenment: “AI agents” are this generation’s “no-code” framework - just like the past generation had CASE, RAD, and, well, no-code.

There were very attractive diagrams shown on the conference. Most of them looked something like this:

[ user input ] -> [ magic ] -> [ business ]

An example of that, and a case I’m interested in professionally, would look like:

[ company name ] -> [ find company description, key people, verified contacts ] -> [ send introductory email ] 

That looks great! I want that agent!

I’m become quite proficient with LLMs, and have done a fair share of agentic workflows, mostly from scratch, but I always felt like I’m missing something. This middle box, the “find company description, key people, verified contacts” one - surely people are using AI agents in some way I’m currently missing out on.

Maybe they created really advanced prompts, that use tools heavily to scrape the web, that find really deeply linked information. Maybe they don’t honour robots.txt? Are they scraping like conference speakers lists and resolve captchas to get to real e-mail addresses? I’d like to to get some pointers to that.

One of the speakers shared his n8n agents with the public, and I jumped at the occasion to see what’s in the box.

Well, it was a REST call to a commercial service that provides that information.

That company presumably does all the actual hard stuff behind the scenes - and fair enough, it’s a viable business - but surely just accessing an API isn’t really “agentic”, right? Agents should have agency, they should decide how to carry out a task, right?

Then it hit me - I’ve actually started noticing the age of the speakers, many of whom were boasting with gathering million+ investments for their startups - and most of them looked under 30.

Re-inventing the wheel is how we get better wheels, it’s a part of the human nature. When I was that age, I too was rather creating stuff from scratch, better and more efficient than the big companies could make as a part of a gigantic product. I crusaded operating systems more obscure than Linux, in a time when Linux was mostly unknown - because I understood them and could work with them, and make them run more efficient than Linux. I’m still a fan of the approach: better build something you understand, than use a black box. Another reason to do it that way is because there’s so much stuff out there to learn. You’d never finish learning if you tried to actually learn everything. So, you arrive at a cut-off point where you just say “ok, I know enough, let’s go build something” - and what you build kind of branches off your own personal tech-tree knowledge, and you figure out a name for it. If it becomes widely accepted, people flock around it and then you have a community which knows the lingo.

AI agents are for people that would rather say:

“Get me a list of verified contacts for this company and put it into our CRM”

than something like:

resp = requests.get("https://api.example.com/get_contacts?company=Company")
if resp.status_code == 200:
   data = resp.json()
   requests.post("http://ourcompanycrm.example.com/save_contact", 
     json={"name": data["company_name"], "address": data["company_address"]})

The first example uses an AI agent, the second doesn’t.

The part that gets lost in the hype is that the agent will most likely use an API service that runs the code similar to the second example - or for simpler cases, it will write the code and run it.

But, isn’t that the holy grail of “IT”? The vast majority of IT is about automation and simplification. Why write a letter by hand and send it by carrier pigeons if you can send an IM / DM?

Every big revolution in technology makes some products or services more available and more accessible to a bigger set of people, even though it introduces new layers of complexity. Blacksmiths of history became material engineers, and we need less of them than before.

The new generation of people productive in IT will not know how to code. Instead, they will prompt LLMs to do stuff. For now, they recognize other members of the same meta-group by using the keyword “AI agents” - that’s a shibboleth. It will one day become obsolete.

Edit: but in the meantime, beware of problems such as these: