Agentic Software: Close Enough for Horse Shoes
November 11, 2025
Another short essay from a real and still-alive dinobaby. If you see an image, we used AI. The dinobaby is not an artist like Grandma Moses.
I read a document that I would describe as tortured. The lingo was trendy. The charts and graphs sported trendy colors. The data gathering seemed to be a mix of “interviews” and other people’s research. Plus the write up was a bit scattered. I prefer the rigidity of old-fashioned organization. Nevertheless, I did spot one chunk of information that I found interesting.
The title of the research report (sort of an MBA- or blue chip consulting firm-type of document) is “State of Agentic AI: Founder’s Edition.” I think it was issued in March 2025, but with backdating popular, who knows. I had the research report in my files, and yesterday (November 3, 2025) I was gathering some background information for a talk I am giving on November 6, 2025. The document walked through data about the use of software to replace people. Actually, the smart software agents generally do several things according to the agent vendors’ marketing collateral. The cited document restated these items this way:
- Agents are set up to reach specific goals
- Agents are used to reason which means “break down their main goal … into smaller manageable tasks and think about the next best steps.”
- Agents operate without any humans in India or Pakistan operating invisibly and behind the scenes
- Agents can consult a “memory” of previous tasks, “experiences,” work, etc.
Agents, when properly set up and trained, can perform about as well as a human. I came away from the tan and pink charts with a ball park figure of 75 to 80 percent reliability. Close enough for horseshoes? Yep.
There is a run down of pricing options. Pricing seems to be challenge for the vendors with API usage charges and traditional software licensing used by a third of the agentic vendors.
Now here’s the most important segment from the document:
We asked founders in our survey: “What are the biggest issues you have encountered when deploying AI Agents for your customers? Please rank them in order of magnitude (e.g. Rank 1 assigned to the biggest issue)” The results of the Top 3 issues were illuminating: we’ve frequently heard that integrating with legacy tech stacks and dealing with data quality issues are painful. These issues haven’t gone away; they’ve merely been eclipsed by other major problems. Namely:
- Difficulties in integrating AI agents into existing customer/company workflows, and the human-agent interface (60% of respondents)
- Employee resistance and non-technical factors (50% of respondents)
- Data privacy and security (50% of respondents).
Here’s the chart tallying the results:
Several ideas crossed my mind as I worked through this research data:
- Getting the human-software interface right is a problem. I know from my work at places like the University of Michigan, the Modern Language Association, and Thomson-Reuters that people have idiosyncratic ways to do their jobs. Two people with similar jobs add the equivalent of extra dashboard lights and yard gnomes to the process. Agentic software at this time is not particularly skilled in the dashboard LED and concrete gnome facets of a work process. Maybe someday, but right now, that’s a common deal breaker. Employees says, “I want my concrete unicorn, thank you.”
- Humans say they are into mobile phones, smart in-car entertainment systems, and customer service systems that do not deliver any customer service whatsoever. Change as somebody from Harvard said in a lecture: “Change is hard.” Yeah, and it may not get any easier if the humanoid thinks he or she will allowed to find their future pushing burritos at the El Nopal Restaurant in the near future.
- Agentic software vendors assume that licensees will allow their creations to suck up corporate data, keep company secrets, and avoid disappointing customers by presenting proprietary information to a competitor. Security is “regular” enterprise software is a bit of a challenge. Security in a new type of agentic software is likely to be the equivalent of a ride on roller coaster which has tossed several middle school kids to their death and cut off the foot of a popular female. She survived, but now has a non-smart, non-human replacement.
Net net: Agentic software will be deployed. Most of its work will be good enough. Why will this be tolerated in personnel, customer service, loan approvals, and similar jobs? The answer is reduced headcounts. Humans cost money to manage. Humans want health care. Humans want raises. Software which is good enough seems to cost less. Therefore, welcome to the agentic future.
Stephen E Arnold, November 11, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
