I have been using AI tools every day for serious technical work for six months. Here is what actually changed and what did not. - has anyone done this

[BretHart99]

Right, six months in. Daily use across coding, research, writing and system design. I want to give an honest account because most of what I read is either cheerleading or backlash and neither maps to my actual experience. The short version is that it changed my workflow significantly but not in the ways I expected going in, and the productivity gains I thought I would get in some areas mostly did not materialise, while genuine improvements showed up somewhere else entirely.

What actually got better: reading and summarising long technical documents, drafting first passes at things I would otherwise procrastinate on, rubber duck debugging when there is nobody around to talk through a problem with, and generating boilerplate I do not want to think about. The METR survey published last week had technical workers reporting a median speed increase of around 3x and value increase of around 2x. I would put myself closer to 1.5x on value and 2x on speed for the specific tasks where AI helps. The number is real but it is not evenly distributed. It is concentrated on the dull stuff.

What did not change: anything requiring genuine domain judgment, anything where the stakes are high enough that I am checking every word anyway, anything involving real institutional context that the model simply does not have. The productivity gains can evaporate if you factor in the time spent on bad outputs, prompt iteration, and the verification overhead. The Deloitte Australia situation where a Big Four firm delivered a report full of AI generated fabrications including fake court cases and non-existent academic references is the horror story version of something I feel as a low grade drag every day.

The thing that surprised me most is a change I was not looking for. My thinking before drafting something has gotten sharper. Because I know I am going to throw a rough version at a model and get structure back quickly, I spend more time on the actual ideas rather than the presentation. The AI absorbed the presentation anxiety and gave me the thinking time back. That is the genuine unlock for me and I have not seen it mentioned anywhere

[HeartbreakKidOscar97]

The last paragraph is the most interesting thing I have read about AI workflow in months. The thinking time reallocation is exactly what happened to me and I had not articulated it that way

[Teal Sparrow]

I had the same realisation around month three. You stop dreading the blank page because the blank page phase is now very short. What expands is the time you spend deciding what you actually want to say

[Luca76]

Counterpoint though. Is that sharpening real or is it just that the lower effort route exists and you are rationalising taking it more often

[ProperMadlad20]

That is a fair challenge but I can test it. The outputs I produce now when I do go through the full manual process are better than they were six months ago. Something transferred

[PlanckLimit81]

Interesting. I have had the opposite experience on that specific point. My unassisted writing has gotten slightly worse because I am less practiced at grinding through the hard part

[Upsilon]

That is a real risk. The skill atrophy problem is underreported. Using a GPS does not improve your map reading ability

[codeberg]

GPS analogy does not quite hold though because writing is not like navigation. The skill you want to retain is the thinking, and if AI is handling the transcription while you keep the thinking, the important part is preserved

[AnthonyCribb]

Agree with that distinction but I am not sure most people are actually using it that way. Most people are using it to skip the thinking as well as the transcription

[Ellie_28]

The verification overhead point is the one that kills me. I work in a context where I cannot afford to let hallucinations through, so I check everything, which means the time saving is much smaller than advertised

[Ava_75]

Same here. I work in regulatory compliance and the checking overhead basically eliminates the speed gain on anything that goes to a client. I still use it for internal drafts but the ROI is not what people claim

[SlowSocket]

For pure coding tasks I still get 2x to 3x on things where the tests confirm correctness. The verification cost drops when the computer can do the verification for you

[SortedBuilder]

That is the underrated point. The domains where AI saves the most time are the ones where ground truth is cheap to check. Code tests, maths, structured data. Natural language is much harder to verify so the gains are smaller in practice

[WovenScholar]

The procrastination unlock is real for me too. I write way more first drafts than I used to because the cost of a bad first draft feels lower. Most of them are still bad but occasionally one becomes something

[EntangledOne]

Lowering the cost of starting is maybe the most durable benefit and I do not see it listed in any of the productivity studies

[Scholar29]

How long before the sheer volume of AI assisted output everywhere starts to devalue clear human thinking. Not asking rhetorically, genuinely wondering what the equilibrium looks like

[Plateau65]

Probably the same way email devalued a well written letter. The form gets commoditised, the signal shifts to something else. In this case maybe original insight or authentic voice

[MrRicardo]

My company rolled out mandatory AI tools six months ago and measured productivity using output volume. Volume went up. Quality of thinking went down measurably in code reviews. Nobody at the top is tracking the right thing

[DigitalNomad76]

Output volume is the laziest possible productivity metric for knowledge work and it was lazy before AI. This just makes it more obviously broken

[Nathan75]

Six months in here too. The thing I miss is the incidental learning that happened when I had to figure something out myself. Reaching for the model first is faster and worse for my development simultaneously

[DeepInlet]

great stuff my friend