@jakevanclief: People are wasting their time building AI agents. A model on its own needs context to be useful: prompts and instructions, tools it can call, and data it can trust. The old way was to write code that wires those three to the model, call it agent one, and do it again for every workflow. Harnesses like Claude Code and Codex already let an AI read and write your files, so you can put your prompts, tools and data in folders and let one AI become whatever agent your question needs. If you want something to study for the next ten years, I'd pick the Unix philosophy, Douglas Engelbart's 1962 report Augmenting Human Intellect, and plain old software logic. The long version, Do You Actually Need AI Agents?, is on my YouTube. Link in my bio. youtube.com/@JEVanClief #ai #aiagents #claudecode

Jake Van Clief
Jake Van Clief
Open In TikTok:
Region: JP
Thursday 22 October 2026 20:02:07 GMT
211132
8618
265
1442

Music

Download

Comments

notatravesty
Wilford Brimless :
DeepSeek harness is my buddy.
2026-10-24 19:54:00
0
joelpelletier7
Joel :
This is why obsidian has changed the game for me.
2026-10-23 13:40:19
7
dadthatbods
Jason Blythe :
I don’t trust anyone who says the “expoDentially” 😂😂🙃
2026-10-22 23:09:54
107
toastedcoconut111
toastedcoconut1 :
what you're describing is just a well designed agent. the next frontier up is an orchestrator that activates these agents by understanding the task at hand and directing them
2026-10-23 02:07:25
76
dawill0w
Will :
But you need agents burning tokens or else how are these AI companies supposed to profit [Tears of joy].
2026-10-24 18:33:31
0
grapedrankyomama
The vacuum that sucks :
There’s this old concept, garbage in garbage out
2026-10-24 18:41:52
0
netouss
Netouss :
Thanks for your args! I can use them in my job, my boss want me to make agent for something, another agent for another thing... But he didn't want to understand It's already super simple to ask a model in my IDE with my workspace in it, and it's super fast because all my IA tools structure is optimized 😂
2026-10-24 20:02:59
0
abdhmmd
Abd :
Agents are just models with it own context, every chat is technically a disposable agent, it is just a terminologie
2026-10-24 20:12:24
0
kennovation
kennovation :
So you’re telling me I’ve been doing it right the entire time and everyone else has just been over complicating things?
2026-10-23 04:29:38
5
ulan.dangalak
Ulan Dangalak! :
So you’re creating an isolated project folder filled with subfolders containing data and code just for that project. Where or how does this “harness” differ from this?
2026-10-23 05:28:44
7
jcost111
jcost111 :
What about governance, user authentication and authorization, OBO tokens from hosts to downstream agents, and being able to do this in a deployed environment where cross-client isolation matters? Wouldn’t the former method be more applicable in such cases?
2026-10-23 01:59:10
10
someonelikemiie
Ella :
Could you explain what you mean by harness?
2026-10-23 03:05:03
5
penkmatters
penkmatters :
Lol, ive been doing this the whole time just because i thought it would be easier
2026-10-22 21:03:24
16
askproai
Quacksquacks :
I also have the best way to understand how you get better to the office by riding a horse to the office
2026-10-23 08:41:37
0
efejotapi
Pino :
Nice take. My company is pushing agents hard, in copilot.
2026-10-22 21:32:18
7
maorbri
Maor :
isn't that what skills are?
2026-10-24 05:26:48
0
freewhil
freewhil :
In other words, for Ai to be useful and accurate you need to have a PHd in prompt writing
2026-10-23 18:00:30
6
neoleeeooo
neo must die :
The scale of data and depth of the workflow makes observability and resilience a nightmare with this style. Long chain context makes agents behaviours subtly different and those differences can make or break certain workflows
2026-10-23 00:21:39
9
hjsanchez_mtz
Héctor Sánchez :
It does work like that, but the idea of having specialized agents is when you need governance, this works for a single person, but once you have a team or in a big company you need governance, so Pablo can do some things and some things can't, but yes I'm with you in the idea
2026-10-22 22:44:47
25
notyourmonk3y
notyourmonk3y :
Ha ha I’ve been doing this for months on my own thinking lol
2026-10-24 15:15:11
2
imnomuse
ImNoMuse :
I have… no clue what you’re talking about. Somehow, the algorithm confused me with someone who knows what any of those words mean. I even watched the whole video twice. 😞
2026-10-23 02:59:08
1
gaffizie0
Godwin Affizie :
this is exactly what I do. until until I started building agents a few days ago
2026-10-23 18:19:30
0
jjamulla
John Jamulla :
Doesn't this use a TON of tokens every time it reads these folders and all the files?
2026-10-23 16:54:12
2
jdmrtl
JD🪐 :
The old way is complicated but you save so much money telling the model what to use, when and how, instead of it having to read it all and find out.
2026-10-23 17:49:39
0
paulheimiller
strongbad :
this is fine for working in your own space. but if developer is writing an agent for an application others will use, you want maximum repeatability and structured flow. in that case using a programming language is necessary.
2026-10-23 10:45:38
3
To see more videos from user @jakevanclief, please go to the Tikwm homepage.

Other Videos


About