I used Claude Code for that PR. I didn’t learn anything. I felt like I was flinging slop over the wall to an open-source maintainer. I felt like a fraud and my impostor syndrome got worse.

But here’s the thing, I still contributed something of value. It’s something I have been wanting in Hugo for a long time to syntax highlight ERB snippets in my posts. It was approved and merged by the maintainer (thanks for dealing with my slop, Alec).

It just feels odd. I know there are many people that are excited about this new era of writing code. But to me this has sucked out all of the fun. I have never felt like a bigger fraud in this field. I’ve always thought that I’m privileged that I get to have fun at all in my line of work. As Ori Bernstein says: “using LLMs to write code is as fun as hiring a taskrabbit to solve my jigsaw puzzles”.

But then again, I know that realistically I would not have the mental capacity or skill to create a pull request like that without AI tooling. My brain is already fried from work on most days. I don’t think I would have been able to learn the codebase and get enough context to make that PR all by myself. It’s a lot. I just wanted some ERB syntax highlighting for my little blog.

Even at work I’ve used Claude Code and other AI tooling to deliver fixes and improvements that have real customer impact. But no matter how big the impact, I feel empty. I agree with Xe Iaso in the first sentence of their post: “Whenever I have Claude do something for me, I feel nothing about the results”.

Now that using AI is a normal expectation at work and how I’m evaluated in performance reviews, I suspect that this fraud feeling will only grow. The industry as a whole is incentivizing delivering code/features/fixes at a quick pace even if it’s all just slop.

I keep thinking a lot about how I perhaps tied my identity too much to my career. I’m not the greatest engineer, but I’ve always worked hard to deliver good work and learn as much as possible. I care a lot about understanding underlying systems as much as possible. I care about the craftsmanship of my code (to the best of my abilities). Unlike me, AI tools don’t care about any of these values.

At the end of the day, the shareholders care about delivering features, gaining customers, and making money. They don’t care how software is built.

I don’t know what to make of this.

Discussion on Hacker News

Basically, it comes down to this: You pass in a string, and you get a string in return. That’s it.

So if we want to run a LLM locally using Python to summarize files, we build strings with Python and pass them into Ollama. If you want to read in files, open them in Python and concatenate the text with your prompt string. Then pass in the prompt string to Ollama.

Python is just a bridge between you and Ollama.

I’ve included some basic examples. The examples assume you have Ollama installed locally.

Reading a Single File Into Ollama

This is straightforward. Open a file and concatenate the text with a prompt which gets passed into Ollama:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

import ollama

# open a single file
file = open("path/to/file.txt")

# read it and concatenate to the prompt
prompt = f'Can you summarize this file for me? {file.read()}'

# pass in the prompt to Ollama
response = ollama.chat(
    model='gpt-oss:20b',
    messages=[
        {
            'role': 'user',
            'content': prompt
        }
    ]
)

print(response.message['content'])

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

import ollama

# open several textfiles
file = open("path/to/file.txt")
file2 = open("path/to/file2.txt")

# concatenate the textfiles into a single string
input = file.read() + file2.read()

# concatenate into the prompt
prompt = f'Can you summarize the following text for me? {input}'

# pass in the prompt to Ollama
response = ollama.chat(
    model='gpt-oss:20b',
    messages=[
        {
            'role': 'user',
            'content': prompt
        }
    ]
)

print(response.message['content'])

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

import ollama

# open several textfiles and store them in a `files` list.
files = []
files.append(open("path/to/file.txt"))
files.append(open("path/to/file2.txt"))

# run a chat for each file in the list
for file in files:
    prompt = f'Can you summarize the following text for me? {file.read()}'

    # pass in the prompt to Ollama
    response = ollama.chat(
        model='gpt-oss:20b',
        messages=[
            {
                'role': 'user',
                'content': prompt
            }
        ]
    )

    print(response.message['content'])

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

import ollama

# save summaries in a list to summarize later on
summaries = []

# open several textfiles and store them in a `files` list.
files = []
files.append(open("path/to/file.txt"))
files.append(open("path/to/file2.txt"))

# run a chat for each file in the list
for file in files:
    prompt = f'Can you summarize the following text for me? {file.read()}'

    # pass in the prompt to Ollama
    response = ollama.chat(
        model='gpt-oss:20b',
        messages=[
            {
                'role': 'user',
                'content': prompt
            }
        ]
    )

    # append the summary of the file to the summaries list for later
    summaries.append(response.message['content'])


# create a single string from our list of summaries
summaries_string = "\n".join(summaries)

# start a final chat to summarize the summaries
prompt = f'Can you summarize the following text for me? {summaries_string}'

response = ollama.chat(
    model='gpt-oss:20b',
    messages=[
        {
            'role': 'user',
            'content': prompt
        }
    ]
)

print(response.message['content'])