Alper Kaan Odabasoglu (@lbc_of_ai) - Twitter Profili

@claudeai I’m sharing some of my Claude usage pipeline through a real product development cycle. You can follow me to get some insights. x.com/lbc_of_ai/stat…

Alper Kaan Odabasoglu@lbc_of_ai

Been improving my app's UI with Claude and found the thing that actually moves the needle. Frontend skill + Claude on its own gets you functional but generic. The jump happens when you feed it real references. I pull a few screens from Dribbble and other pro work, drop them in, and the output goes from "fine" to something I'd actually ship. The model isn't the bottleneck. Your taste input is. Most people skip the reference step and blame the tool. Before and after below.

English

0

1

29

Claude@claudeai·15h

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

English

6K

6.8K

66.9K

17.1M

Alper Kaan Odabasoglu@lbc_of_ai·14h

@batodabasoglu @claudeai 😂😂😂

QME

0

1

41

Batuhan Odabasoglu@batodabasoglu·14h

@lbc_of_ai @claudeai It all starts with "yo check this website" and you see the link that starts with localhost

English

1

0

1

25

Alper Kaan Odabasoglu@lbc_of_ai·14h

Claude@claudeai

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

ZXX

1

0

2

108

Alper Kaan Odabasoglu@lbc_of_ai·14h

@claudeai

QME

0

14

1.3K

Alper Kaan Odabasoglu@lbc_of_ai·14h

them: fable 5 access extended til july 19, go build something incredible the incredible things: localhost:3000 localhost:5173 localhost:8080

Claude@claudeai

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

English

0

1

50

Alper Kaan Odabasoglu@lbc_of_ai·14h

@claudeai fable 5 doing my planning and orchestration has been the actual unlock for me, everything downstream routes off it. extra week of access means i dont have to think about swapping my whole pipeline mid-build.

English

0

2

1.1K

Alper Kaan Odabasoglu@lbc_of_ai·18h

wrote a code verification skill for my memory app where the checklist is the single source of truth one markdown file holds every convention. two things apply it: - pre-push git hook (blocks push until it runs) - a PR workflow that splits the checklist into ~6 dimensions, one Opus subagent each, fused into one report both read the same file so they can't drift curious how others handle this: one file both the hook and the PR check read, or separate configs per tool?

English

0

1

2

75

Alper Kaan Odabasoglu@lbc_of_ai·1d

@batodabasoglu Awesome…

English

0

1

51

Batuhan Odabasoglu@batodabasoglu·1d

This is Fulcrum. Months of building this in silence, no name, no audience, just the work. Testing ideas and killing most of them, because they didn't survive scrutiny, and scrutiny was the whole game. Think of it as a system, not a single signal. A core engine that stays invested and rotates with the market's regime. A turbo bolted onto that engine, extra power that kicks in only when the conditions call for it, quiet the rest of the time. And around both, a set of smaller satellites reading volume and pressure, the stuff most people scroll past. Those satellites are becoming the freemium layer of what's next: market insight and technical reads, open to anyone who wants to follow along before any of it becomes a subscription. Backtested CAGR (2005-2026): 21.4% vs QQQ's 14.9%. A -43.1% max drawdown sits right next to that number, on purpose. Still in paper trading. Backtested/hypothetical performance, shared for informational purposes only. Not investment advice. Everyone talks the talk. Watch this walk. I'll be sharing the rest of the process here, stick around.

English

1

2

3

460

Alper Kaan Odabasoglu@lbc_of_ai·2d

@DanielSmidstrup Check this post :) x.com/lbc_of_ai/stat…

Alper Kaan Odabasoglu@lbc_of_ai

Been improving my app's UI with Claude and found the thing that actually moves the needle. Frontend skill + Claude on its own gets you functional but generic. The jump happens when you feed it real references. I pull a few screens from Dribbble and other pro work, drop them in, and the output goes from "fine" to something I'd actually ship. The model isn't the bottleneck. Your taste input is. Most people skip the reference step and blame the tool. Before and after below.

English

0

2

26

Daniel Smidstrup@DanielSmidstrup·2d

Tell me one thing you can do that CLAUDE cannot do yet

English

259

2

125

15.6K

Alper Kaan Odabasoglu@lbc_of_ai·2d

@ClaudeCodeLog Check this post :) x.com/lbc_of_ai/stat…

Alper Kaan Odabasoglu@lbc_of_ai

Been improving my app's UI with Claude and found the thing that actually moves the needle. Frontend skill + Claude on its own gets you functional but generic. The jump happens when you feed it real references. I pull a few screens from Dribbble and other pro work, drop them in, and the output goes from "fine" to something I'd actually ship. The model isn't the bottleneck. Your taste input is. Most people skip the reference step and blame the tool. Before and after below.

English

0

1

35

Claude Code Changelog@ClaudeCodeLog·3d

Claude Code 2.1.206 has been released. 27 CLI changes Highlights: • Security help restricted to authorized/educational use; DoS, supply-chain and evasion requests are refused • Background agents auto-upgrade after Claude Code updates to avoid slow stale-session upgrades on attach • Expired logins now prompt to run /login instead of a misleading model error, showing correct auth guidance Complete details in thread ↓

English

9

18

322

58.6K

Alper Kaan Odabasoglu@lbc_of_ai·2d

@ClaudeDevs x.com/lbc_of_ai/stat… Check this post to get some insights :)

Alper Kaan Odabasoglu@lbc_of_ai

Been improving my app's UI with Claude and found the thing that actually moves the needle. Frontend skill + Claude on its own gets you functional but generic. The jump happens when you feed it real references. I pull a few screens from Dribbble and other pro work, drop them in, and the output goes from "fine" to something I'd actually ship. The model isn't the bottleneck. Your taste input is. Most people skip the reference step and blame the tool. Before and after below.

English

0

1

21

ClaudeDevs@ClaudeDevs·2d

Claude Code on desktop now has an in-app browser. Claude can pull up docs, designs, or any other site. It can read, click through, and interact the same way it does with your local dev servers. It's sandboxed and configurable: you choose whether sessions persist.

English

701

1.5K

21.7K

3.2M

Alper Kaan Odabasoglu@lbc_of_ai·2d

If this helped and you know someone building with Claude Code, a repost would mean a lot.

English

0

3

64

Alper Kaan Odabasoglu@lbc_of_ai·2d

Been improving my app's UI with Claude and found the thing that actually moves the needle. Frontend skill + Claude on its own gets you functional but generic. The jump happens when you feed it real references. I pull a few screens from Dribbble and other pro work, drop them in, and the output goes from "fine" to something I'd actually ship. The model isn't the bottleneck. Your taste input is. Most people skip the reference step and blame the tool. Before and after below.

English

4

7

12

376

Alper Kaan Odabasoglu@lbc_of_ai·2d

Imagination plus the right references plus AI. That combo is the actual unlock right now. You bring the taste and the direction, Claude does the building, and the loop is fast enough that the only real limit is how well you can picture what you want. I post all of it here. The pipeline as I refine it, model routing and Claude Code setups and verification tricks, and the products I'm building with it. Follow along if that's your thing. Let's build. (or until I hit my Claude session limit lol)

English

0

3

64

Alper Kaan Odabasoglu@lbc_of_ai·2d

Step one is planning, and I use Fable 5 for it. The trick is I put the visual direction in the plan itself. The Dribbble references, the token mapping, the card layout, all of it goes into the plan so the execution model isn't making design calls. Then Opus or Sonnet agents just follow the plan. Taste doesn't survive a model handoff unless you write it down. (the prompt in this screenshot is a mock, just to show how I structure it, not the actual one I used)

English

1

0

3

97

Alper Kaan Odabasoglu@lbc_of_ai·3d

@ClaudeDevs You guys are awesome :)

English

0

15

ClaudeDevs@ClaudeDevs·3d

We've reset 5-hour and weekly rate limits for all users.

English

3.6K

3.1K

48K

7.2M

Alper Kaan Odabasoglu@lbc_of_ai·5d

Concrete version from my build: in plan mode, Fable 5 specs the whole feature before a line of code exists. Then Opus 4.8 or Sonnet 5 executes depending on complexity, working against a plan it didn't have to invent mid-stream. Rework dropped. You can layer loop engineering on top with a goal condition, but be careful with your stop conditions and scope. Easy to burn significant amount of token budget in one run if you're not watching it. :D

English

1

0

3

1.3K

ClaudeDevs@ClaudeDevs·5d

A few patterns we frequently use with Fable 5: Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate.