It's insane that you can get a blood test in Riga for 10 eur, it took me 5 minutes, gave blood at 6pm, and at 10pm got the result. Private blood testing in London is a fucking joke
how are you designing dynamodb table?
DynamoDB is great tool, but it's not optimal if you don't all of your access patterns for the data, i.e. adding extra access patterns later might be painful. Just FYI.
If you don't know all access patterns and might want to change fields, change access patterns - RDS is a safer choice
I agree that the build speed has increased now.
But, LOC alone is not a great metric. nobody wants 100k LOC of slop per day.
I've seen agentic tools confidently create code and tests for features that nobody asked for and later removed. Also, generating extra code requires more code traversal to understand what's going on (even by ai agents), this just polutes context.
In my workflow it's more nuanced than 50/50:
How critical is the code change (i.e. if it touches auth, payments - full review).
Most changes - glance through it, sanity check, make sure tests are there.
Any small feature implementations (with no new patterns/architecture changes, and existing tests) - YOLO ship it.
Sure, happy to help.
Looks like feedback button works with adblock disabled.
For the guide on animation - yep, I've seen it, but it's for your internal pixel engine model, I'm referreing to the character generation which uses nano banana models. There was no prompt enchancement option.
There were a few subtle UX bugs/incosistencies but nothing major.
`Feedback` button doesn't work so couldn't report these :)
- The middle area is always dark, not sure what should go there, but there is so much wasted space (screenshot)
- You can click on remix on a gif, but you get an error "pixel-engine-1.1 requires a pixel art image."
- Reframe only works on a single image not a series of frames
- Character generation - can't generate characters with transparent background (guess it's coming form nano banana)
- There is an issue with displaying a border in asset selection popup (screenshot)
Also, would be good to have a guide on how to prompt characters. I guess there is some system prompt that you pass to nano banana, but would be good to see what works and what doesn't. It took me a lot of generations to start getting decent results.
Btw, if you plan to build a way to generate multiple assets the same time, check recraft - they had a good way of generating a few assets the same time (in explore mode it works great), they have a canvas where you can generate 4+ assets and select the one you like the most.
Have you tried https://t.co/kF255YO0Jw ?
I had good results with it, but I generated a bunch of docs before the initial prompts, so all prompts have same context.
You can also use MCP to generate screens, but would recommend automating it with MCP only after you got the reference screen right.
A few days ago I've asked for good marketing focused CLAUDE.md examples. Still haven't found it, but found a great repo for marketing skills - https://t.co/lankb258Bd by @coreyhainesco
In early stages of a product you want to prioritise the speed of iteration.
DynamoDB focuses on access patterns first. But while working on an MVP you might not know all access patterns in advance.
Postgres makes it easier to iterate in early stages of a product. Effortless scaling for early users also isn't a problem with postgres.
I'm building a tool (@DynoWizard) that helps to design single table DynamoDB databases, but even that tool is using postgres under the hood.
Changed my workflow when using LLM based tools.
Now the first step is research:
Opus 4.6 creates a doc on how to get the most of a specific tool (best practices, what to avoid, examples, etc), then reference the doc before using the tool.
Increases the quality of generations.
Some of them are now dedicated skills in Claude Code, but all of them are in a dedicated repo "RTFM"
@helloitsolly why not have the marketing style page for engineering role? I'm software eng and the marketing style page for hiring just shows the company and branding identity
🧵 In 2004, Amazon's shopping carts went down during the holiday rush. Databases saturated. Real revenue, gone per second.
That failure started an 8-year path to DynamoDB. The path was weirder than most people realize:
Depends how you approach it. I actually find it can work great, but you still have to craft eng strategy or have good documentation on how the software should work. i.e. you need to have constraints and guide tobwhat you want. And you can't add features non stop, in a green project you'd still have to stop and reevaluate tech decisions and refactor code.