Tom

Tom

@imtomdev.bsky.social

Software engineer Building a SaaS by day, building my home by night πŸ—οΈ Testing AI coding tools on camera: aidive.dev Β· youtube.com/@AIDive-YT

774 followers95 following247 posts

Pinned

The part I did not expect: the security scanner it ships with was last published in March. Six months behind the repo it audits.

AI Dive@aidive.bsky.social

Fresh install of everything-claude-code (250k stars), then I measured what it puts in context before you type: about 27k tokens, 14% of a 200k window, every session. Mostly descriptions of 286 skills you will never call. Five parts are worth copying out. www.youtube.com/watch?v=iu9-...

The cost line is not the part that stuck with me. It's the 5,488 words of artifacts somebody on the team has to read.

AI Dive@aidive.bsky.social

I ran Anthropic's AI-native SDLC playbook end to end on one repo, 16 timed sessions, $12 total. Same bug fixed directly: 2:13, 70 cents. Through all six gates: 11:31, $3.46. Three of the six pay that back. www.youtube.com/watch?v=bL08...

The part I did not expect: the same /verify, run by Opus instead of the model that wrote the code, failed that change 3 out of 3.

AI Dive@aidive.bsky.social

I ran 116 Claude Code sessions on a real repo and graded every "done" with tests the agent never saw. The built-in /verify came back PASS 24 times out of 24, cost 2.5x more on Opus, and changed no outcome. www.youtube.com/watch?v=oL2g...

The part I did not expect: the file told the model to import from an alias that does not exist. No run ever used it.

AI Dive@aidive.bsky.social

I deleted the 177-line CLAUDE.md from a real repo and ran the same five tasks 44 times. One rule broke: on a new file, 3 of 4 runs hardcoded a string instead of adding the translation. Every other rule held, the code next door already taught them. www.youtube.com/watch?v=1oJS...

The zero I did not expect: three MCP servers installed, 63 runs, not one tool call.

AI Dive@aidive.bsky.social

I installed the eight Claude Code repos that top every must-have list onto one real project and ran the same tasks with and without each. All eight cost under 7k tokens at session start. The three MCP servers they ship were called zero times in 63 runs. www.youtube.com/watch?v=73Sz...

The part that stuck with me: a Mod missing a capability fails closed, but a Mod that's just slow gets skipped and the command runs anyway.

AI Dive@aidive.bsky.social

I wrote a 42-line Claude Code Mod that redacts secrets from tool output before the model reads them. 28 ms per call, zero tokens. Then I put a 15 second sleep in it: after 10 s the engine skipped the guard and ran the command anyway. www.youtube.com/watch?v=YGxV...

The line I keep thinking about: the OpenSpec bake-off spec run cost three times as much and still found three gaps the plain run missed.

AI Dive@aidive.bsky.social

okama's maintainers found eight agent-written plans rendered as public Read the Docs pages, committed exactly as Superpowers intends. Where specs go is solved. Who owns them and when one is dead is not, so I wrote down five rules for that. www.youtube.com/watch?v=IKfU...

the number that surprised me: only one shell command in three even has an rtk rule. the dashboard looks great, the bill barely notices. the tool that cut my usage most was the one nobody installs for tokens

AI Dive@aidive.bsky.social

JetBrains found rtk made Claude Code sessions 7.6% more expensive. My rtk dashboard says 11.6M tokens saved. Both are right: the dashboard counts shell bytes, not the bill. I ranked the four token savers I run by what they move: www.youtube.com/watch?v=Dees...

The bit I keep coming back to: Anthropic's own support page says Fable spends your allowance faster. That is written down, not a rumour.

AI Dive@aidive.bsky.social

Fable 5.1 cache reads are $0.25 per million. Opus 5's are $0.50. That is the only line in the pricing where the expensive model is the cheap one, and it only shows up if your agent keeps re-reading the same transcript. www.youtube.com/watch?v=azLx...

Yesterday, I struggled with loading and empty states in my account list. Making both work together means juggling local state in React, causing re-renders 🧡

If you're using my extension and this bothers you, let me know! Would show me users care about it πŸ˜‚ For now, focusing on the landing page!

Found a way to skip testing my app: use it daily πŸ˜… But seriously, building a product you use yourself makes it easier to spot critical paths to test

We're close to BluePeers 1.0.0! Just launched a major update: activity score πŸ“ˆ Key features: β€’ Score from 0 to 100 based on last 100 interactions β€’ Includes posts & replies (excludes reposts) Goal: help you find active peers and keep that amazing interaction vibe we've had this past month πŸ¦‹

When people report issues with my product, I feel this urgency to fix them ASAP - don't want to lose users. But bugs and feedback are normal, especially when building in public. Guess I need to get used to it

View full thread

The clipboard feature is another game-changer. More complete than Alfred's, with content types and extra info. Best part? OCR search in images. Finding specific images used to need separate tools, now it's built right in. That seamless integration is 🀌

Raycast doesn't have a workflow builder like Alfred, but lets you use scripts (bash, applescript...). While I don't use it much yet, it's promising - especially with AI. Just tell Cursor or ChatGPT what you want, and boom: custom script ready.

Raycast's store has amazing integrations. As a French dev writing in English, the Chrome Translate integration is perfect - everything happens right in Raycast with a smooth UX.

I have to improve my landing page. Thinking about showcasing BluePeers features with videos/GIFs - each feature gets a GIF and description. Considering a tab system: one tab = one feature = one GIF. WDYT? Would this be cool? #buildinpublic

New branding for BluePeers ✨ Just shipped a fresh release and updated the Chrome store visuals. Not a designer, but I think it looks better than before! #buildinpublic

Just launched a new release for BluePeers! - Added like/unlike account filtering - Fixed some bugs with existing filters Also refreshed the Chrome visuals to look more professional, I'll tell you more on another post πŸ‘€ bluepeers.app #buildinpublic

I love the #buildinpublic vibe. In the past, I hesitated to share my thoughts and my journey. Today, with Bluesky's positive vibe, I finally took the leap and I'm not disappointed at all. I'm getting new ideas and feeling more motivated. BTW: thanks @fabiulous.work for sharing your ideas

I'm working on introducing authenticated call on my Chrome extension Why ? I want to add some extra features that requires auth calls. And I'm encountering some difficulties 🧡 #buildinpublic

View full thread

I don't have to deal with third party. But I have to inform users about the process : - How to create an app password : url to the settings - Gif tutorial : showing the action of creating an app password and using it in my app - Warning about not using their main Bluesky password

I'm working on introducing authenticated call on my Chrome extension Why ? I want to add some extra features that requires auth calls. And I'm encountering some difficulties 🧡 #buildinpublic

View full thread

Because each app password can be deleted at any time. Compared to login/password, the barrier is thinner and I think I'll lose fewer people at the start.

I'm working on introducing authenticated call on my Chrome extension Why ? I want to add some extra features that requires auth calls. And I'm encountering some difficulties 🧡 #buildinpublic

View full thread

I thought about connecting user through a classic email/password. But I want a seamless auth method and this one creates a barrier to entry. Also, I want to make Bluesky authenticated calls so random email/password isn't needed. If I ask email/password and then Bluesky handle, it's a lack of time…

I saw a feature that I didn't know about: app password. It seems to fit perfectly. I just want to perform Bluesky auth calls and app password give temporary access to that. Why temporary ?

I'm working on introducing authenticated call on my Chrome extension Why ? I want to add some extra features that requires auth calls. And I'm encountering some difficulties 🧡 #buildinpublic

At first, I wanted to add OAuth to my app but I'm facing some issues. The documentation about OAuth is very dense and it seems that OAuth only works for web apps. With google extension, it's not quite simple : I have to deal with OAuth callback URL, and at this time, it seems too complicated

Every. Single. Time: - Navigate your app naturally, you know it by heart - Start recording your launch video - Completely forget how your app works Anyway πŸ˜… New release out: filters added! bluepeers.app

Currently trying to maximize the context of Cursor by adding a lot of comments in my files. Of course, I don't add the comments myself; I ask Cursor to do it, and I hope that after many iterations, the AI will have a better understanding of my app. I put my prompt in the first comment πŸ‘‡

The prompt ➑️ I would like you to add comments to this file. You can proceed in two different ways. Either: - comment code using comments - comment code by renaming variables (code as documentation) Important: I'm only asking you to add comments, NEVER modify the code structure

To easily find my peers, I want to filter by following/followers, so I decided to include some filters in my extension. They will be saved in Chrome storage to persist across sessions. Release submitted soon πŸ‘€ #buildinpublic

Been very uncomfortable about pitching my SaaS to future English-speaking clients. As a French person, the language barrier feels scary

Last night, my extension was approved by Google, so it's launch time! πŸš€ bluepeers.app A free tool to help you find active peer accounts directly inside πŸ¦‹. See interesting data about accounts you come across on the interface: listings, feed, notifications Would love your feedback πŸ™

Sometimes I hesitate sharing dev fails or life stuff here. Brain's like "maybe keep that one to yourself" but screw it - the realest convos always start when someone decides to be honest first

Spent way too much time overthinking tech choices. Now I just pick what lets me ship fastest. If it works and doesn't slow me down, that's good enough. The rest is just noise