By the end of 2026, how useful do you think agentic/multi-agent AI systems will actually be in your daily work or personal projects?
Displaying poll results.3229 total votes.
Most Votes
- By the end of 2026, how useful do you think agentic/multi-agent AI systems will actually be in your daily work or personal projects? Posted on March 11th, 2026 | 3229 votes
Most Comments
- By the end of 2026, how useful do you think agentic/multi-agent AI systems will actually be in your daily work or personal projects? Posted on March 11th, 2026 | 27 comments
Hallucinations are most dangerous when rareish (Score:3)
The more common the hallucinations are, the less dangerous they become because everyone begins to expect them.
They are most dangerous when they are just rare enough to be unexpected.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Just because the local drunkard becomes violent every evening after his drinks doesn't mean you can become used to him.
Um... real world evidence seems to contradict this. Ex. The unhoused people in various cities. Another example: the weather (ex. cities that often get tons of snow versus cities where it hardly ever snows) - people get used to it when it's frequent, but lose their shit if it's uncommon.
Ask someone who lives above a dive bar if they get used to the scene. (hint: they do)
Re:Hallucinations are most dangerous when rareish (Score:4, Insightful)
The more common the hallucinations are, the less dangerous they become because everyone begins to expect them.
We are talking about autonomously acting AI agents here, not just about a chatbot.
If such an agent "hallucinates" that you need two truckloads of ping pong balls and places an order to be delivered at your front door, you don't want that to become common.
vscreen (Score:1)
Re: (Score:2)
What do scented urinal mats have to do with this?
Is it 2027 and this poll is still active? (Score:5, Funny)
Missing option: "It's already 2027, you insensitive clod!"
The previous Slashdot poll stayed active for something like 10 to 12 months. I miss the old days when we got a new poll every week!
Re: (Score:2)
Same. The polls were also there to start a conversation and not just market research.
Hypotheticals for 2027? (Score:2)
Thanks for clarifying the joke. I didn't get it at first.
The poll needs a Cowboy Neal option, but that might be his obituary in 2027?
Me? I had to answer hypothetically because I'm retired. I had to go for the first answer though I would prefer to be a refusenik. I can absolutely see how using the current genAI tools would be extremely powerful for most of the work I did before I retired. What that means is I would be delivering less for my salary if I refused to use the AI tools. Even worse than that, someo
Re: (Score:2)
Re: (Score:2)
These days the companies don't want any expensive humans doing that sort of expensive support. Therefore it doesn't matter if the AIs can't provide it.
My "latest and greatest" theory is that the leading corporate cancers have decided it is better to push expensive customers away in the hopes that they will reduce the relative profitability of the surviving competitors. It doesn't even matter if they are profitable customers as long as the related profits are relatively smaller than the profits from "better"
Re: Hypotheticals for 2027? (Score:2)
Three weeks ago I shifted to using AI for a minor coding task. This week I converted 30GB of geodata from 5 sources into a mountain layer for a game map at 10 meter resolution. Today I used it full time to compare 16 laws to an architecture document to see if I missed something.
3 months ago none of this was really easy. This month it found 4 python bugs in a library I was using, created a fix, tested the fix, committed it, synced it and drafted the PR (which I checked and yes the whole thing was good).
I'm a
Re: (Score:2)
Thanks for the evidence, but it still makes me sad.
Re: (Score:1)
I switched camps about a year ago and have found that they are like any tool, you have to know how to use it right without cutting yourself or shooting your feet off.
Another critical factor is that they can be both garbage and amazing at the same time. Sadly most of us are using hosted online services for these and the quality can vary by the day. You have to be a
Hallucinations et al. (Score:1)
In my daily life, I don't see a lot of people being aware of hallucinating AI agents. Some skepticism, some raving "oh my god this is amazing!" but I don't hear a lot of "cautiousness about results," so I've been a bit more vocal when appropriate.
That being said, there are plenty of edge case uses where AI agents completely fail. I'm still in the middle of testing the usefulness of Claude, ChatGPT and Gemini in a specific situation. I'm trying to see if it can help me decide which characters in a mobile g
Recent experience (Score:3)
Up until very recently, I'd only used ChatGPT as a front end to a search engine and to produce examples of single-line PowerShell.
Then I tried something more complex that required iterative improvements and discovered that AI session tokens don't have any chronological context. The longer a session, the more likely it will introduce things you'd already removed, no matter how emphatically you instruct the AI not to do that.
I was completely unaware that adding a time factor and discarding unhelpful results was difficult to build into an AI.
Re: (Score:2)
I voted game changer because I've noticed something similar in my recent usage. This is probably old hat for experts in AI, but I've noticed that the agentic AI brags about having millions of tokens worth of context. I think that context is what makes some AIs better at not getting into those idiot loops where you tell it not to do it one way, it thinks, then later it does it the wrong way again. I've compared the big guys with the little models you can run on your own GPUs and the context seems to matter t
Re: Recent experience (Score:2)
Context windows make a big difference both in how well it can keep a straight line and also in how much memory and processing power you need. On my PC with 24GB GPU and 64GB ddr5 I can run reasonably effective models (up to 12GB) with a 60K context window. That's good enough for small stuff.
Codex 5.3 can do 230K, can even go much higher when using the API, and auto compacts the token window. That makes a big difference.
What also helps is to maintain a task list, tell it to split up any assignment and put it
They're better, but still have flaws (Score:2)
They've gotten better in recent months.
I just hate that I mostly use it the same way I used to use google before. Google now tends to prioritize AI written garbage. The AI, ironically, can often dig up pages with actual research.
Most of which will vaguely summarize a topic, but have no actual research on whatever topic you're looking for. If you don't really constrain it, the ai will tend to go on tangents and generate lots of fluff.
You're always going to end up running into hallucinations if whatever you'r
No use what so ever (Score:3)
They won't be any use at all.
I work in the physical world, not the digital world.
sometimes (Score:1)
My productivity is up 5x. Minimum. (Score:2)
I have multiple AI services in my recently purchased Jetbrains Junie subscription. Right now I'm using codex for my work. It feels like having a pro developer team of 10 API experts at your service and ready to answer any request on your code base or any complex question right away. I've already taken the back seat and let AI do must of the hard work as per my instructions andrequests.
It's absolutely amazing. I've been professionally doing webdev for 25 years now and never had this competent a colleague, me
Re:My productivity is up 5x. Minimum. (Score:4, Funny)
With a 5x productivity boost, I hope you're enjoying the 5x pay rise your employer must obviously be giving you, as a reward for the 5x more profit you're generating for them! Good on you!
Specific tasks, but sporadic usage (Score:1)
- a way to "find one word and replace it randomly from a bank of other options without 2 of the same word replaced in succession," or
- obscure calculator ideas that I wonder about, like "If Atari's Superman game costs $27.99 on a 4k ROM in 1978, how much would that cost-per-KB rate translate to a modern game KB/MB/GB size with an equivalent modern dollar" that don't have any real impa
VSCode Copilot does pretty good with common tasks (Score:2)
I've lately been using Copilot in VSCode to build a custom CRM/ticket management app to replace an ancient thing that was built in the early 2000s, in COBOL, and has various conniptions running on anything newer than XP.
I'd say the agent has autonomously written.. probably 95% of the code for a FastAPI backend and Vue.js frontend. Mostly with the model selection set on Auto. It seems to mainly use Codex and Claude Sonnet, delegating lighter tasks to Haiku. I can manually select some of the free models if
I think I hallucinated this (Score:2)
Considering how many errors AI makes with even very simple factual queries, anyone using them professionally is running a massive risk. I mean, I have yet to ask an AI a legal question and get even a remotely accurate response. Most of the time it's either wrong, incomplete or actually just made up.