Which AI agents do you still use every week, and which ones faded out?

Real habits, not benchmarks — here's mine as a solo dev shipping a small SaaS:
Claude is my daily driver, but the thing that actually changed my workflow wasn't the model itself, it was using it for stuff that isn't coding. I've been leaning on it hard for the un-fun side of being solo reading my own GA4 and Search Console data, figuring out where traffic actually comes from, deciding what to do next. Turns out "talk through my analytics with something that pushes back" is a more valuable habit than "generate code faster." I didn't expect that.
GPT I keep around for quick one-off questions and when I want a second opinion that isn't going to agree with the first one. Different failure modes, so cross-checking is useful. What faded for me: the autonomous-agent stuff. I tried a few "let it run and do the whole task" setups during the hype and they were impressive in demos but I never built a habit around them , too much babysitting to trust on anything that touches real users or data. I came back to the boring loop of me + one assistant in a tight feedback cycle.
On your routing point I think you're onto something real, but the honest friction for me isn't "which model is best for this task," it's that I don't want to think about routing at all. The work day it picks for me invisibly and I never see the seam is the day it becomes a habit instead of a decision. Right now choosing still feels like work, and anything that feels like work is what fades.

I'm curious what your test users say about that — do they actually want to compare models, or do they want to stop having to?

The agents that have stuck for me are the ones I can name as part of an operating loop, not just a tool: triage this inbox, summarize this customer thread, turn this repo context into a patch, check this run before I ship.

The ones that fade are usually impressive in demos but need a fresh decision every time: what should I ask, where should I paste context, how do I verify it? If the handoff and verification are still manual, the habit never really compounds.

the ones that stick usually remove a decision, not just save keystrokes.

I use Claude for anything that needs real thinking, GPT for quick answers, Cursor for code. The tools that faded were the ones that made me context-switch too much, habit forms around whatever removes friction fastest.

Claude (via API for my product + via Claude Code for shipping it), Cursor for fast IDE work, Perplexity for research. The one that faded out: any general-purpose agent that promised to do everything. The ones that stuck were either specialized (one job, done well) or extensible enough to become specialized in my workflow.

The routing instinct is the right one, the agents that stick are the ones with a job clear enough that you know exactly when to reach for them. The ones that faded for me were the general do-anything agents, because deciding what to ask them was more work than just doing the task myself.

Claude is my daily driver for anything large-context. The 1M window is genuinely great for holding a whole codebase or doc set at once. For quick stuff I switch to Gemini Flash. I haven't done much rigorous experimentation though, it just felt right a few months ago and got baked into the stack.

That's the part I find hard. These model choices get locked in with almost no re-evaluation. Swapping is technically easy, but I never actually go back to check whether the choice still holds after a few model releases. A lightweight way to re-score models against your own real workflows would be great. Not sure where I'd even start building that one.

Is that something EvoLink is thinking about, or is it more the routing/access layer?

The stickiness question is the one worth asking. For me the tools that stuck all have one thing in common: they fit into an existing motion rather than requiring a new one.

Claude is my main tool for anything that needs real thinking — writing, planning, longer research. Cursor for coding. Perplexity for quick fact checks. Those three have survived every round of new launches because they slot into how I already work.

The ones that faded: I tried 4-5 different AI writing tools over the past year. All had good demos. None made it past week 3. The pattern was always the same — I had to change my workflow to use them rather than them fitting into my existing one.

I built Genie 007 partly because of this exact frustration — voice input that catches you wherever you are without switching apps. The tools that become habits remove friction from behaviour you already do.

The routing question you're raising is real. Different models do perform differently on specific task types and most people stick with one sub-optimally because switching cost is too high. What's EvoLink's routing logic — prompt classification, task type, or latency/cost optimisation?

Answering your actual questions: Claude for anything long, structured, or code, ChatGPT for fast thinking and search-style queries, and honestly that's most of it week to week. What faded for me wasn't the weak tools, it was good ones that lived in a separate tab. The cost nobody prices in is context. Every new agent makes you re-explain your project, your constraints, your voice, so the tool that wins is whichever one you've already loaded all that into. That's why switching feels expensive even when the new option is technically better. On the routing angle you're circling, my contrarian take after running AI inside a real product: most operators don't have a routing problem, they have a too-many-tools problem. One strong default, plus routing only the two workflows where cost or quality genuinely diverges, beats a clever router you then have to babysit.

For me, ChatGPT and Claude are the ones that stuck. Not necessarily because they're always the "best," but because they've become part of my daily workflow. Most of the AI tools I stopped using weren't bad—they just never solved a recurring problem well enough to become a habit. I think adoption is less about benchmarks and more about whether a tool consistently saves time on tasks you already do every day.

Interesting post, I was just discussing this with a friend last night

Which AI agent or assistant do you actually use every week? - Claude and OpenAi

Do you use different agents for different tasks? I tend to default to ChatGPT/Codex combo but I'll often ask the same thing of claude at the same time and start to balance out the responses.

What tasks do you still prefer Claude, GPT, Gemini, Grok, Cursor, Codex, or other agents for? ChatGPT for marketing for image generation for sure

Which tools did you try during the hype, but eventually stop using? Grok and Perplexity. Grok was more 'fun' for a while but I felt it lacked the depth for serious work. Perplexity just felt weak at everything. Copilot I tried a few times but junk

the ones that faded weren't bad. they just never got embedded in a workflow that already existed. the tools that stuck are the ones that replaced something i was already doing somewhere else, not ones that required me to build a new habit from scratch

Right now, I only use Codex. It lets me work on projects and chat in the same interface, and it now supports connecting Windows with iOS, which means I can use Codex from anywhere. As for why I don’t use CC, it’s because it has strict restrictions in my region.

For me, ChatGPT is still the one I use consistently.
I've tried several AI tools over the past few months, but most of them were interesting for a week and then disappeared from my workflow.
The tools that survive are usually the ones that save time every single day.

Built my entire SaaS with AI coding assistance. The ones that stuck:

• Claude for architecture decisions and debugging
• Cursor for daily coding (the autocomplete is addictive)
  The ones that faded:
• Most "AI agent" frameworks — too much overhead for solo founders
• Anything that requires more setup than the problem it solves
  Lesson: the best AI tool is the one you actually use, not the one with the most features.

Claude Desktop is the one I keep coming back to for actual work. The others I tried (various browser agents, AutoGPT-style tools) were fun for a week then disappeared from my workflow.

The biggest friction I hit with Claude is that I run multiple sessions simultaneously and they constantly pause on permission prompts — especially when I step away. That problem has eaten more of my productivity than any other single thing.

Curious if others have that issue or if you're mostly running single-session workflows?

I went all-in on Claude for coding a few months ago and barely touch ChatGPT now, but I still keep Perplexity around for quick research stuff — feels like it earned a permanent tab in my browser.

I NEED A TRUSTED CRYPTO HACKER THAT CAN RESTORE LOST OR SCAMMED FUNDS.

Are you struggling to get back the money you lost? Every day, countless individuals face the devastating impact of scam operations that drain their hard-earned savings. But there’s good news – GEO COORDINATES RECOVERY HACKER are here to help you recover what’s rightfully yours. I lost my entire savings to a fake crypto investment scam while I was looking for a way to double my savings. After many weeks of trying to find a way to get my money back with no success, I finally came across a crypto recovery company GEO COORDINATES RECOVERY HACKER, a reliable and trustworthy crypto recovery company. I'm immensely grateful for his dedication, professionalism, and unwavering support. You can get in touch with them through below contact details

WhatsApp ; +1 ( 318 ) 203-3657

I had to send out my review also. They are indeed recommendable.

The agents that stick for me are the ones with a repeatable loop, not the ones with the best demo. Claude/Codex for coding when I can keep the context tight, GPT for quick research and first-pass writing, and smaller models only when latency/cost matters more than judgment.

The ones that faded are usually the “agent that does everything” tools. If I cannot predict what it will touch, how much context it will burn, or how easy it is to review the diff/output, it becomes a novelty instead of a habit.

One practical filter: I trust an agent more when I can name the unit of work and measure the session afterward. “Fix this auth edge case” survives. “Go improve the app” usually turns into context soup.

The point ibiti made about habits is the real insight, tools stick when they plug into a repeatable workflow. One part of the AI stack I still think gets overlooked is the input layer. Everyone is busy optimizing which model to send things to, but the typing bottleneck between thought and prompt is real. Dictating prompts instead of typing them cuts filler naturally, speaking tends to strip out the waffle that typing encourages. I built DictaFlow for exactly this: hold-to-talk voice dictation that types into any app. dictaflow.io

I use AI Manus, Claude and GPT to achieve a balance result. I certainly use different agents for different tasks.
I still prefer using Claude, Manus and GPT for my daily task because I get best results every time.

I’ve noticed something similar on the dev side — I don’t really stick to one agent anymore, it’s more like different tools for different parts of the workflow (debugging, structuring logic, quick implementation ideas).

Curious if others are also moving from “one main tool” to a more task-based setup.

The “habit” point is the strongest part here.

Most AI tools do not fade because the model is bad. They fade because they never attach to a repeatable workflow. Once a tool becomes part of writing, coding, research, support, or data work, switching becomes harder.

For EvoLink, I’d be careful not to frame it mainly as “access and compare models through one API.” That sounds useful, but still a bit infrastructure-generic.

The sharper angle might be closer to workflow routing: helping teams decide which model should handle which job, based on cost, speed, quality, context length, and reliability.

That makes the product feel less like another model gateway and more like the decision layer behind AI workflows.

Good question. My rotation has settled to three, and the routing between them became the actual workflow pattern:

Claude stays for production code and architecture reasoning. When I need to audit a codebase for a specific pattern or debug something with deep context, Claude is the only one I trust for the real work.

ChatGPT got added back for the quick stuff — rewriting email drafts, brainstorming landing page copy, generating 10 variations of a subject line. The latency improvement made it viable for throwaway tasks that would waste Claude's context budget.

Perplexity is the surprise keeper for research queries. When I need to know the actual state of something like automated testing patterns in Astro or how WebGPU rendering works in different browsers, having inline citations saves the copy-paste-search loop.

The ones that faded: all the specialized wrapper tools that were a single model behind a thin UI. Devin, various agent platforms. They tried to own the whole workflow but the switching cost between their opinionated environment and just opening Claude directly was never worth it.

I'm still routing manually. At team scale I can see the appeal of a unified API, but for solo work the overhead of a router layer hasn't justified itself yet.