I have tried various AI code assistants in the past (Copilot, Aider, etc.), but nothing really stuck. Recently, I have been using Claude code some. This tool runs in the terminal and seems like a better fit for my workflows, which are centered around the terminal, instead of an editor like VS Code. So far, I’m impressed with what it can do – anything from adding a feature to some Elm code to proofreading a blog post.
Claude code also offers a very reasonable pay-as-you-go model, which is a good fit for trying it out or occasional use.
Has anyone else tried Claude code? What do you think of it?
Claude proofreading seems like a nice balance – it seems to catch the obvious spelling and grammar errors, but is not quite as overbearing as Grammarly.
Claude models are named after poems, where the names reflect the length of the poem:
Opus (most powerful)
Sonnet (balances capability and efficiency)
Haiku (optimized for speed and affordability)
It is interesting in the above proofreading example, it appears Claude code uses a mixture of models for different tasks.
Term
Basic Definition (Writing Context)
Opus
A substantial and significant work, typically referring to a major piece of literature, music, or art. In writing, it often means an author’s most important or ambitious creation—sometimes called a “magnum opus” when referring to a writer’s greatest achievement.
Sonnet
A poem consisting of 14 lines written in iambic pentameter, following a specific rhyme scheme and thematic structure. Sonnets are known for their strict form and are used to explore complex ideas within a concise format.
Haiku
A short poem of Japanese origin, consisting of 17 syllables in three lines of 5, 7, and 5 syllables respectively. Haikus focus on imagery and often capture a moment in nature or daily life.
Experimenting with using Claude code to add some custom shell commands and create a debugging stepping through a state machine. It needed some correction, but 30 or so states, it probably save some time typing. Interestingly, it did not use the Opus module, so I might manually switch that that next time:
Interview with two senior devs about using AI for coding.
tl; dr: AI works great for senior devs and things you already know how to do (makes you much faster). It does NOT work very well for things you don’t know how to do.
I had a little personal side project where I needed to OCR a bunch of faded typewritten scanned pages from PDF and translate them from German to English. Google lens was the only tool I found that could do the OCR effectively on the faded documents. I then wondered – does Google have an an API for Lens? Sure enough: Google Cloud Vision API.
I asked Claude to generate a program in Go that takes a directory of images, OCR’s them using the vision API, and then translates them. It took Claude about 10m to write the program and debug the compile errors. I then needed to set up a credentials file – setting up GCP or AWS AIM is not my idea of a good time, so I asked Claude to use gcloud in my terminal and set up an AIM profile – success, it created a credentials file that worked:
I then asked it to process a PDF directly by splitting it into separate images for each page – it took a few iterations, but it got it.
It probably took another hour or so of prompting and testing to get the output text formatted like I wanted it.
Cost was about $5 to write the program using Sonnet model. Sonnet is working really well for simple tasks like this, so will be intersting to see what Opus can do.
can you combine the 5 PDFs in this directory into a single PDF using the tools available on this system?
● I’ll help you combine the 5 PDFs into a single PDF. Let me first check what PDF files are in the directory and
then use available tools to combine them.
● List(.)
⎿ Listed 9 paths (ctrl+r to expand)
● Now I’ll combine the 5 PDFs using a PDF merge tool. I’ll use pdfunite which is commonly available on Linux
systems:
● Bash(pdfunite “Main_Fig1_PFC_boost-Auxiliary_source_circuit_2025-Mar-15.pdf”
“Main_Fig2_DC-BUCK_batt_chg_circuit_2025-Mar-15.pdf” "Main_Fig3_DC-BOOST_5VDC_aux_source…)
⎿ (No content)
● Perfect! I’ve successfully combined all 5 PDFs into a single file called
Combined_Power_Supply_Inverter_Diagrams.pdf. The combined PDF contains all the circuit diagrams in order:
PFC boost auxiliary source circuit
DC-BUCK battery charging circuit
DC-BOOST 5VDC auxiliary source
DC-AC inverter microcontroller connections
Control panel connections
The combined PDF is now available in your current directory.
Claude is really good at writing scripts, macros, etc. Recently had it write a Libreoffice Python macro – is is really good at that.
Claude seems pretty effective in Yocto projects. One of the challenges with Yocto is processing large numbers of files and summarizing what is going on – this is something AI is good at.
must be a little too popular, as I see this occasionally…
> add a user login page
⎿ API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}
> add a user login page
⎿ API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}
> add a user login page
⎿ API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}
> add a user login page
⎿ API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}
> add a user login page
⎿ API Error: 500 {"type":"error","error":{"type":"api_error","message":"Overloaded"}}
The feature is available in Claude Code’s own terminal/editor and can be turned on at any time by running /vim. You may also use /config to adjust these settings.
Mode switching: Esc (to NORMAL mode), i, I, a, A, o, O (to INSERT mode)
Instead of building one-off Jupyter notebooks that get discarded, the team now has Claude build permanent React dashboards that can be reused across future model evaluations.
This is perhaps one of the most interesting uses of AI coding – helping to build YOUR Platform. Typically, these tasks are not very hard, but time-consuming.
Another thing that stands out in this post is they use Claude a lot for learning.
There are multiple mentions of using Markdown and CSV files – I have come to the same conclusion. In the past, these simple formats were harder to work with, but with AI tools, this is no longer the case. You don’t need Excel to create a graph from a table of data – with a few commands, you can now have some reusable code that does this. You could probably build an entire financial app using a bunch of CSV files and some Claude commands/rules (ClaudeFinance?). Or perhaps a PLM workflow … ClaudePLM anyone?
Other quotes that stood out:
While supervising, don’t hesitate to stop Claude and ask “why are you doing this? Try something simpler.” The model tends toward more complex solutions by default but responds well to requests for simpler approaches.
They frequently use screenshots to show Claude Code what they want interfaces to look like, then iterate based on visual feedback rather than describing features in text.
The investment required to build out this tech is staggering:
Six months after DeepSeek Monday, Anthropic is looking to scale up yet again. The company is in talks for a new funding round that could reach $5 billion which would double its valuation to $150 billion. The potential investors include some middle eastern Gulf states that Anthropic once seemed eager to steer clear of. But after raising nearly $20 billion from Google, Amazon, and VCs like Lightspeed, it’s running out of options for bigger checks.
Within Anthropic, Amodei has argued the Gulf states have $100 billion or more in capital to invest, and their cash would help Anthropic stay on the technology’s frontier. He seemed to reluctantly accept the idea of taking the money from dictators in an internal Slack message obtained by Wired. “Unfortunately,” he wrote. “I think ‘No bad person should ever benefit from our success’ is a pretty difficult principle to run a business on.”
