Comment by msejas
8 hours ago
Most people don't understand that 'agentic AI' is a completely synthetic, force fed capability by extensive Reinforcement Learning on synthetic domain specific 'agentic' datasets on post training.
If the LLM wasn't post-trained to adhere to specific handbook, it just won't work. If the LLM wasn't trained on an use case the lab decided was worth making a synthetic agentic dataset, it won't work as well as you want.
There's a reason the main agentic task LLMs excel at are coding tasks, it's the way of working of the creators, and they understand intimately the flow and can train for it.
I believe the true way will be able to easily fine tune models on your agentic use cases, but it would require a big company to compile a huge dataset on it's way of working and I don't think anyone wants to be the first.
In terms of long context, accurate attention retrieval from early tokens is just impossible, given the expansion of RoPE encoding for the positions, or in case of Kimi that don't use it anymore, as well as deepseek, early context is heavily compressed you lose accurate information.
If people spent more time studying about AI and how it works, they would realize that the default should be to one shot prompt your task with a big, cached system prmopt, with an user prompt that is just dynamic data, specified to the cheapest model that can do the job.
Unless you really can't do this given your problem, you should try to make a graph of well defined, step by step oneshot prompts, and THEN if your problem still can't be solved with that, then you start leveraging agents.
Despite this giving better results, and being more cost efficient, is evidently too much work then just letting the AI do all the work.
What do you mean by a graph of one shot prompts?
Most people jump straight to agents when what they actually need is a graph. Example: a mining company receives free-text reports from field geologists. You could have:
Geologist report -> LLM call extracts minerals we are looking for (you inject a db query result on the user prompt), locations, assay mentions and risks into structured fields -> LLM call classifies evidence into positive indicators, negative indicators and unknowns -> LLM call estimates deposit potential and confidence -> database lookups inject regional ore demand, nearby deposits, infrastructure and historical yield data -> LLM call combines geological evidence with business context -> LLM call generates an investment recommendation and rationale.
That's what I mean by graph. Every step is a separate LLM call with a well-defined responsibility, consuming the output of the previous node. Each node can be tested, benchmarked, retrained, replaced, or monitored independently. Why would you use an agent here? You can cache every single system prompt on each call making your total token output much cheaper than having a full 'output' only token generation workflow which is what happens with agents.
There is nothing to discover. The workflow is already known. The company already knows how geologists evaluate prospects. The company already knows what data sources matter. The company already knows what the final output should look like. You don't want the model deciding which tools to call, which reasoning path to take, or which pieces of information are important every single run. You want the exact same process applied to every report so results are consistent, measurable, auditable and debuggable. My default is: One-shot prompt -> if not enough -> graph of LLM calls -> if not enough -> agent. A surprising amount of enterprise AI is really just: Unstructured input -> extraction -> classification -> enrichment from databases -> decision support. Not: Unstructured input -> autonomous agent spends 20 steps deciding what to do next.
Agents make sense when the workflow itself is unknown.
If the workflow is already understood, a graph is usually cheaper, more reliable, easier to evaluate, easier to debug, and less dependent on whatever synthetic "agentic" behaviors happened to get reinforced during post-training. I am sure people default to agents mostly because it's less engineering work than explicitly modeling the process.
Thanks, that helped. I get it. Yea, that’s the harness executing the workflow with the LLM being called at the right time. That ensures the process is consistent no matter what, in contrast to the agent calling out to tools and possibly doing different things every time. Basically, standard code being in control rather than the LLM being in control. Fully agree with this model. We should use deterministic code when we want the same process or algorithm every time and choose LLM callouts when we want “fuzzy” processing that is not as deterministic.
Isn’t this just software? But with an LLM at various points instead of a deterministic call to some function?
This is a great point, and one that has been blowing my mind for a while. The public think of AI as this black box thing approaching human intelligence. They don't know that when you look under the hood of a lot of ai products you see a string of prompts that anyone could put together themselves. The companies are just putting them together in a workflow.
1 reply →
I would call that a pipeline, which may be a type of graph, but might have communicated your point more concretely to begin with
Can I have graphs as the default and fall back to agentic behavior if the prompt / task can't be mapped to a known process graph?
1 reply →
I jump straight to graphs (unless it truly cannot be solved with a graph), but then the stakeholders get upset that I didn't use agents. Doesn't fit their marketing plan, I suppose. I don't mind dealing with that, but I can see why most jump to agents in order to avoid the human conflict.
> In terms of long context, accurate attention retrieval from early tokens is just impossible, given the expansion of RoPE encoding for the positions, or in case of Kimi that don't use it anymore, as well as deepseek, early context is heavily compressed you lose accurate information.
"Every gambler knows the secret to survivin' is knowing what to throw away, and knowing what to keep." - Kenny Rogers
Humans have limited context, just like AI. The difference is that humans - at least some of the time - can figure out which pieces are more likely to be important, and therefore prioritize keeping those in the context.
can downvoters explain? this has been by experience with these tools too.
i thought it was well known that claude code got good at coding because anthropic bought tons of coding data from companies like mercor.