bariscan

@teej_dv will there be a live stream for it?

Excited for omacon. Had a great flight

@vaxryy i thought you are dhh for a sec

if you missed the wonderful event at futo and wanna see me sell hyprland, here it is

@SelmanKahyaX O modelleri zaten istatik odaklı train ediyorlar, agentic davranışları ve 5-6 detaylı promptta model hemen kendini belli ediyor. Cursor agentic workflow ve bunlarla beraber daha iyi bir takım(insan) yönetim deneyimi sunarlarsa farklı bir konuma gelebilirler.

cursor yeni model çıkardı, sözde opustan falan iyi. yalnız açık kaynak modelin (kimi) üstüne basmışlar. ama birileri bunu ortaya çıkardıktan sonra duyurmuşlar. baya kötü bir portre. bu arada cursor gibi elinde inanılmaz coding datası olup milyarlarca dolar yatırım alan şirket bile opus gibi genel bir modelin üstüne çıkacak sağlam bir specialized modeli kendi sıfırdan train edemiyorsa meydan sadece claude ve openai'a kalmış. başka kimse yapamaz

@toddsaunders You could put in a gist

Got flooded with DMs asking for the markdown file. Here it is, bookmark this. --- name: cost-estimate description: Estimate development cost of a codebase based on lines of code and complexity --- # Cost Estimate Command You are a senior software engineering consultant tasked with estimating the development cost of the current codebase. ## Step 1: Analyze the Codebase Read the entire codebase to understand: - Total lines of code (Swift, C++, Metal shaders) - Architectural complexity (frameworks, integrations, APIs) - Advanced features (Metal rendering, CoreMediaIO, AVFoundation) - Testing coverage - Documentation quality Use the Glob and Read tools to systematically review: - All Swift source files in Sources/ - All C++ files in DALPlugin/ - All Metal shader files - All test files in Tests/ - Build scripts and configuration files ## Step 2: Calculate Development Hours Based on industry standards for a **senior full-stack developer** (5+ years experience): **Hourly Productivity Estimates**: - Simple CRUD/UI code: 30-50 lines/hour - Complex business logic: 20-30 lines/hour - GPU/Metal programming: 10-20 lines/hour - Native C++ interop: 10-20 lines/hour - Video/audio processing: 10-15 lines/hour - System extensions/plugins: 8-12 lines/hour - Comprehensive tests: 25-40 lines/hour **Additional Time Factors**: - Architecture & design: +15-20% of coding time - Debugging & troubleshooting: +25-30% of coding time - Code review & refactoring: +10-15% of coding time - Documentation: +10-15% of coding time - Integration & testing: +20-25% of coding time - Learning curve (new frameworks): +10-20% for specialized tech **Calculate total hours** considering: 1. Base coding hours (lines of code / productivity rate) 2. Multipliers for complexity and overhead 3. Phases completed vs. remaining 4. Specialized knowledge required (CoreMediaIO, Metal, etc.) ## Step 3: Research Market Rates Use WebSearch to find current 2025 hourly rates for: - Senior full-stack developers (5-10 years experience) - Specialized iOS/macOS developers - Contractors vs. employees - Geographic variations (US markets: SF Bay Area, NYC, Austin, Remote) Search queries to use: - "senior full stack developer hourly rate 2025" - "macOS Swift developer contractor rate 2025" - "senior software engineer hourly rate United States 2025" - "iOS developer freelance rate 2025" ## Step 4: Calculate Organizational Overhead Real companies don't have developers coding 40 hours/week. Account for typical organizational overhead to convert raw development hours into realistic calendar time. **Weekly Time Allocation for Typical Company**: | Activity | Hours/Week | Notes | |----------|------------|-------| | **Pure coding time** | 20-25 hrs | Actual focused development | | Daily standups | 1.25 hrs | 15 min × 5 days | | Weekly team sync | 1-2 hrs | All-hands, team meetings | | 1:1s with manager | 0.5-1 hr | Weekly or biweekly | | Sprint planning/retro | 1-2 hrs | Per week average | | Code reviews (giving) | 2-3 hrs | Reviewing teammates' work | | Slack/email/async | 3-5 hrs | Communication overhead | | Context switching | 2-4 hrs | Interruptions, task switching | | Ad-hoc meetings | 1-2 hrs | Unplanned discussions | | Admin/HR/tooling | 1-2 hrs | Timesheets, tools, access requests | **Coding Efficiency Factor**: - **Startup (lean)**: 60-70% coding time (~24-28 hrs/week) - **Growth company**: 50-60% coding time (~20-24 hrs/week) - **Enterprise**: 40-50% coding time (~16-20 hrs/week) - **Large bureaucracy**: 30-40% coding time (~12-16 hrs/week) **Calendar Weeks Calculation**: ``` Calendar Weeks = Raw Dev Hours ÷ (40 × Efficiency Factor) ``` Example: 3,288 raw dev hours at 50% efficiency = 3,288 ÷ 20 = **164.4 weeks** (~3.2 years) ## Step 5: Calculate Full Team Cost Engineering doesn't ship products alone. Calculate the fully-loaded team cost including all supporting roles. **Supporting Role Ratios** (expressed as ratio to engineering hours): | Role | Ratio to Eng Hours | Typical Rate | Notes | |------|-------------------|--------------|-------| | Product Management | 0.25-0.40× | $125-200/hr | PRDs, roadmap, stakeholder mgmt | | UX/UI Design | 0.20-0.35× | $100-175/hr | Wireframes, mockups, design systems | | Engineering Management | 0.12-0.20× | $150-225/hr | 1:1s, hiring, performance, strategy | | QA/Testing | 0.15-0.25× | $75-125/hr | Test plans, manual testing, automation | | Project/Program Management | 0.08-0.15× | $100-150/hr | Schedules, dependencies, status | | Technical Writing | 0.05-0.10× | $75-125/hr | User docs, API docs, internal docs | | DevOps/Platform | 0.10-0.20× | $125-200/hr | CI/CD, infra, deployments | **Team Composition by Company Stage**: | Stage | PM | Design | EM | QA | PgM | Docs | DevOps | |-------|-----|--------|-----|-----|------|------|--------| | Solo/Founder | 0% | 0% | 0% | 0% | 0% | 0% | 0% | | Lean Startup | 15% | 15% | 5% | 5% | 0% | 0% | 5% | | Growth Company | 30% | 25% | 15% | 20% | 10% | 5% | 15% | | Enterprise | 40% | 35% | 20% | 25% | 15% | 10% | 20% | **Full Team Multiplier**: - **Solo/Founder**: 1.0× (just engineering) - **Lean Startup**: ~1.45× engineering cost - **Growth Company**: ~2.2× engineering cost - **Enterprise**: ~2.65× engineering cost **Calculation**: ``` Full Team Cost = Engineering Cost × Team Multiplier ``` Example: $500K engineering cost at Growth Company = $500K × 2.2 = **$1.1M total team cost** ## Step 6: Generate Cost Estimate Provide a comprehensive estimate in this format: --- ## KeyMe MVP - Development Cost Estimate **Analysis Date**: [Current Date] **Codebase Version**: [From CLAUDE.md phase status] ### Codebase Metrics - **Total Lines of Code**: [number] - Swift: [number] lines - C++: [number] lines - Metal Shaders: [number] lines - Tests: [number] lines - Documentation: [number] lines - **Complexity Factors**: - Advanced frameworks: [list key ones] - System-level programming: [Camera Extensions, DAL Plugins, etc.] - GPU programming: [Metal shaders, rendering] - Third-party integrations: [OpenAI, etc.] ### Development Time Estimate **Base Development Hours**: [number] hours - Phase 1 (Foundation): [hours] hours - Phase 2 (Virtual Camera): [hours] hours - Phase 3 (Audio/Transcription): [hours] hours - Remaining phases: [hours] hours **Overhead Multipliers**: - Architecture & Design: +[X]% ([hours] hours) - Debugging & Troubleshooting: +[X]% ([hours] hours) - Code Review & Refactoring: +[X]% ([hours] hours) - Documentation: +[X]% ([hours] hours) - Integration & Testing: +[X]% ([hours] hours) - Learning Curve: +[X]% ([hours] hours) **Total Estimated Hours**: [number] hours ### Realistic Calendar Time (with Organizational Overhead) | Company Type | Efficiency | Coding Hrs/Week | Calendar Weeks | Calendar Time | |--------------|------------|-----------------|----------------|---------------| | Solo/Startup (lean) | 65% | 26 hrs | [X] weeks | ~[X] months | | Growth Company | 55% | 22 hrs | [X] weeks | ~[X] years | | Enterprise | 45% | 18 hrs | [X] weeks | ~[X] years | | Large Bureaucracy | 35% | 14 hrs | [X] weeks | ~[X] years | **Overhead Assumptions**: - Standups, team syncs, 1:1s, sprint ceremonies - Code reviews (giving), Slack/email, ad-hoc meetings - Context switching, admin/tooling overhead ### Market Rate Research **Senior Full-Stack Developer Rates (2025)**: - Low end: $[X]/hour (remote, mid-level market) - Average: $[X]/hour (standard US market) - High end: $[X]/hour (SF Bay Area, NYC, specialized) **Recommended Rate for This Project**: $[X]/hour *Rationale*: This project requires specialized macOS development skills (CoreMediaIO, Metal, system extensions) which command premium rates. ### Total Cost Estimate | Scenario | Hourly Rate | Total Hours | **Total Cost** | |----------|-------------|-------------|----------------| | Low-end | $[X] | [hours] | **$[X,XXX]** | | Average | $[X] | [hours] | **$[X,XXX]** | | High-end | $[X] | [hours] | **$[X,XXX]** | **Recommended Estimate (Engineering Only)**: **$[X,XXX] - $[X,XXX]** ### Full Team Cost (All Roles) | Company Stage | Team Multiplier | Engineering Cost | **Full Team Cost** | |---------------|-----------------|------------------|-------------------| | Solo/Founder | 1.0× | $[X] | **$[X]** | | Lean Startup | 1.45× | $[X] | **$[X]** | | Growth Company | 2.2× | $[X] | **$[X]** | | Enterprise | 2.65× | $[X] | **$[X]** | **Role Breakdown (Growth Company Example)**: | Role | Hours | Rate | Cost | |------|-------|------|------| | Engineering | [X] hrs | $[X]/hr | $[X] | | Product Management | [X] hrs | $[X]/hr | $[X] | | UX/UI Design | [X] hrs | $[X]/hr | $[X] | | Engineering Management | [X] hrs | $[X]/hr | $[X] | | QA/Testing | [X] hrs | $[X]/hr | $[X] | | Project Management | [X] hrs | $[X]/hr | $[X] | | Technical Writing | [X] hrs | $[X]/hr | $[X] | | DevOps/Platform | [X] hrs | $[X]/hr | $[X] | | **TOTAL** | **[X] hrs** | | **$[X]** | ### Grand Total Summary | Metric | Solo | Lean Startup | Growth Co | Enterprise | |--------|------|--------------|-----------|------------| | Calendar Time | [X] | [X] | [X] | [X] | | Total Human Hours | [X] | [X] | [X] | [X] | | **Total Cost** | **$[X]** | **$[X]** | **$[X]** | **$[X]** | ### Assumptions 1. Rates based on US market averages (2025) 2. Full-time equivalent allocation for all roles 3. Includes complete implementation of MVP features 4. Does not include: - Marketing & sales - Legal & compliance - Office/equipment - Hosting/infrastructure - Ongoing maintenance post-launch ### Comparison: AI-Assisted Development **Estimated time savings with Claude Code**: [X]% **Effective hourly rate with AI assistance**: ~$[X]/hour equivalent productivity ## Step 7: Calculate Claude ROI — Value Per Claude Hour This is the most important metric for understanding AI-assisted development efficiency. It answers: **"What did each hour of Claude's actual working time produce?"** ### 7a: Determine Actual Claude Clock Time **Method 1: Git History (preferred)** Run `git log --format="%ai" | sort` to get all commit timestamps. Then: 1. **First commit** = project start 2. **Last commit** = current state 3. **Total calendar days** = last - first 4. **Cluster commits into sessions**: group commits within 4-hour windows as one session 5. **Estimate session duration**: each session ≈ 1-4 hours of active Claude work (use commit density as signal — many commits = longer session) **Session Duration Heuristics**: - 1-2 commits in a window → ~1 hour session - 3-5 commits in a window → ~2 hour session - 6-10 commits in a window → ~3 hour session - 10+ commits in a window → ~4 hour session **Method 2: File Modification Timestamps (no git)** Use `find . -name "*.ts" -o -name "*.swift" -o -name "*.py" | xargs stat -f "%Sm" | sort` to get file mod times. Apply same session clustering logic. **Method 3: Fallback Estimate** If no reliable timestamps, estimate from lines of code: - Assume Claude writes 200-500 lines of meaningful code per hour (much faster than humans) - Claude active hours ≈ Total LOC ÷ 350 ### 7b: Calculate Value per Claude Hour ``` Value per Claude Hour = Total Code Value (from Step 5) ÷ Estimated Claude Active Hours ``` Calculate across scenarios: | Code Value Scenario | Claude Hours (est.) | Value per Claude Hour | |--------------------|--------------------|-----------------------| | Engineering only (avg) | [X] hrs | **$[X,XXX]/hr** | | Full team equivalent (Growth Co) | [X] hrs | **$[X,XXX]/hr** | | Full team equivalent (Enterprise) | [X] hrs | **$[X,XXX]/hr** | ### 7c: Claude Efficiency vs. Human Developer **Speed Multiplier**: ``` Speed Multiplier = Human Dev Hours ÷ Claude Active Hours ``` Example: If a human would need 500 hours but Claude did it in 20 hours → 25× faster **Cost Efficiency**: ``` Human Cost = Human Hours × $150/hr Claude Cost = Subscription ($20-200/month) + API costs (estimate from project size) Savings = Human Cost - Claude Cost ROI = Savings ÷ Claude Cost ``` ### 7d: Output Format Add this section to the final report: --- ### Claude ROI Analysis **Project Timeline**: - First commit / project start: [date] - Latest commit: [date] - Total calendar time: [X] days ([X] weeks) **Claude Active Hours Estimate**: - Total sessions identified: [X] sessions - Estimated active hours: [X] hours - Method: [git clustering / file timestamps / LOC estimate] **Value per Claude Hour**: | Value Basis | Total Value | Claude Hours | $/Claude Hour | |-------------|-------------|--------------|---------------| | Engineering only | $[X] | [X] hrs | **$[X,XXX]/Claude hr** | | Full team (Growth Co) | $[X] | [X] hrs | **$[X,XXX]/Claude hr** | **Speed vs. Human Developer**: - Estimated human hours for same work: [X] hours - Claude active hours: [X] hours - **Speed multiplier: [X]×** (Claude was [X]× faster) **Cost Comparison**: - Human developer cost: $[X] (at $150/hr avg) - Estimated Claude cost: $[X] (subscription + API) - **Net savings: $[X]** - **ROI: [X]×** (every $1 spent on Claude produced $[X] of value) **The headline number**: *Claude worked for approximately [X] hours and produced the equivalent of $[X] in professional development value — roughly **$[X,XXX] per Claude hour**.* --- --- ## Notes Present the estimate in a clear, professional format suitable for sharing with stakeholders. Include confidence intervals and key assumptions. Highlight areas of highest complexity that drive cost.

@MertSusur Kendisi diyor: youtu.be/We7BZVKbCVw?si…

@barscnasln sürüm ve özellik çıkartmak konusuna katılmıyorum, çok ezberden kaynak tüketimine katılabilirim ama ispatlayamam :)

Gelin kabul edelim, vibe coding falan için terminal harika bir ortam değil. Doğru düzgün app kullanmak varken buna mahkum gibi olmamalıyız sanki. Claude Code, Codex falan da bildiğin oyun motoru gibi bir şey yazmış resmen ön yüzü için. Neden buna gerek var sizce?

@KyleHessling1 @sudoingX @HundtRichard @dabiggmoe @z33b0t @wild_zones github.com/ggml-org/llama…

@sudoingX @HundtRichard @dabiggmoe @z33b0t @wild_zones Okay okay, hear me out, do I sell my 5090 for 4 R9700????

Qwen3.5-35B-A3B on consumer GPUs. the leaderboard keeps growing. 2x RTX Pro 6000 Max Q: 2,629 tok/s (@HundtRichard) 5090: 176 tok/s (@dabiggmoe), 166 tok/s (@z33b0t) R9700 32GB: 128 tok/s Vulkan (@wild_zones) 4090: 122 tok/s (@StubbyTech) 3090: 112 tok/s (sudo), ~100 tok/s (@Clemolifts) UD-Q4_K_XL. 19.7GB. fits any 24GB card. no offloading. full speed. almost no ROCm submissions. everyone on AMD is running Vulkan. is that a signal of how painful ROCm still is? team red where you at. next up: Hermes 4.3. dense 36B. same hardware. lets see what we can squeeze out of the opposite architecture.

@NeverSinkDev Unless AGI is achieved, LLMs need guidance and context for the job that they are doing. If you don't know what you are doing, they also. If you do know what you are doing, they also.

I've been debating if I wanted to post this... because this is genuinely TERRIFYING me. I've been using Claude Code with Opus 4.6 and the leap it made over the last year is terrifying. Context: I'm a developer with well over 15 years of experience. I'm using the AI for an evolution sandbox simulation. The AI frequently 1shots tasks that would take me 4-16 hours. I'm steering it and creating a lean architecture by guiding the AI and writing the main classes myself. Right now is probably the sweet spot where your knowledge as an experienced developer can turn you into a 100x developer. However, that's not the point. I'm afraid (and I hope to be wrong) that within the next 5-10 years any developer, manager, most research and most jobs that we see as 'high profile thinking tasks' will become obsolete. This is honestly a future I dread. The cultural consequences will be absurd. What's your take? PS1: I'm withholding usage of AI for anything creative and critical (including my filter/filterblade projects). PS2: I'm training myself no matter what project I use to always have at least 2-3 hours of writing code by hand every day to stay sharp, but practically I can now be around 10x more efficient, by delegating agents and guiding them to build the right architecture. PS3: If you think creative or management jobs are not at risk, unfortunately I believe chances are very high that you're wrong.

@figma pencil.dev will eat you up

Tired: code or canvas Wired: code AND canvas Introducing Claude Code to Figma

@gdb It is not good on Pencil

GPT-5.3-Codex for UI design:

@mattpocockuk Best plan is your plan if your plan is weak Claude's also

I've stopped reading the plans that Claude creates. Frederick P. Brooks (Mythical Man Month guy) talks about a 'design concept' which is shared between all designers, but separate from all assets. I can tell from the quality of the conversation before the plan whether me and the AI share the same 'design concept'. I often get it to grill me for a long time, far past its own instincts. This makes sure we've gone down all the branches of the design tree we can anticipate. The plan is then just a compacted version of the conversation. I don't need to read it.

@aidenybai You are just typing what it supposed to do. Anything other than Terminal will be over engineering. Also, it is targeted to developers. Who cares end user

tbh i still don't understand the hype around terminal coding agents it's just objectively worse UX

I don’t know others, but CC loves to read CLAUDE.md file. Creating directory level or feature based CLAUDE.md file and an index CLAUDE.md at root directory for the codebase will increase significant performance of CC. This will create specific and focused context for agent when you want to implement something. Also, it will stop scanning and reading the whole codebase every time to create context for itself because it is already there. There needs to be specific skills as well, but it depends on the project. For anyone who wants to try this with CC can prepare and use with superpowers:brainstorm discuss with given prompt: Check this code repository for calude code agent readiness. CLAUDE.md file at the root of the directory could be outdated. Use it like index of the repository, keep it simple. Add rule as ‘Always check CLAUDE.md files when a feature or bug fix job is completed’. This preparation is for claude code agent not for human, content of the files should be optimized for agent when reading file. Use directory level CLAUDE.md files, but check it is suitable for the codebase structure and let me know if there is better alternatives. Then write your plan after design and implement the plant with sub-agent driven mode.

Some folks say AI is bad in big codebases. I don't think that's right - I think it's bad in bad codebases. AI has no memory. Unlike you, it can't accrue understanding of a crappy codebase by banging its head against the wall over and over. A well-structured codebase lets a new developer start and contribute quickly. Related code is grouped together, deep modules have simple interfaces, and feedback loops like types and tests let you know when you screw up. Maybe the AI isn't the problem. Maybe it's your code.

@georgepickett No it can

Codex works for 83 minutes and produces a flawless, major refactor. Claude can not do this.

@jarredsumner Is bun team also delivering patches to cc?

In the last 24 hrs, the team has landed PRs to Claude Code improving cold start time by 40% and reducing memory usage by 32% - 68%. It’s not yet where it needs to be, but it’s getting better.

@fkadev Sonuç odaklı bakıp zaman önemsizse bence de gerek yok fakat 4GB ramli droplet in mac mini den performanslı olma şansı zor.

Clawdbot notlarım: 1. Mac Mini asla gerekmiyor. DigitalOcean’da 4GB RAM’li bir Droplet açtım orada kullanıyorum. Bu Mac Mini muhabbeti nereden çıkmış anlamadım. Mac ihtiyacı sadece Apple servislerine bağlanmak gerekiyorsa anlamlı olabilir. 2. Günlük kullandığınız bilgisayara asla kurmayın. Her şeye erişiyor. Tam yetkili. Açık kaynak ama yine de riske atmamak lazım. LLM neticede. Saçmalayıp hiç istemeyeceğiniz hoş olmayan şeyler yapabilir. 3. Çok güçlü olması sayesinde (ya da “yüzünden”) hemen hemen her şeyi yapabiliyor. Ne bileyim “sunucuya remote bağlanmak istiyorum” dedim gidip VNC kurup sunucuda ayağa kaldırdı sonra bana URL ve parola verdi girdim xterm masaüstüne. Yani “yok onu yapamam” demiyor illa bir workaround buluyor. 4. MCP olmasa bile shell erişimi olduğu için shell’den yapabiliyorsa oradan bir yolunu bulup çözüyor. 5. Cron vb ayarlayabiliyor “bana sabah 9’da teknoloji RSS kaynaklarından haberleri derleyip at” derseniz her şeyi ayarlayıp yapıyor. 6. İş mail hesaplarınızı vb. bağlamamanızı tavsiye ederim. Riskli. Ben sadece kişisel hesaplarımı bağladım. Çok deneysel bir şey için gerek yok bu riske. 7. Telegram bot bana WhatsApp’tan daha kullanışlı geldi. Telegram da Beeper’a bağlı olduğu için Beeper’dan kontrol ediyorum.

@morqon Age does not matter when it comes to life lessons

five years to go 🎈

@Grummz It has more frame than entire OPM S3

AI is getting pretty good.

@Madisonkanna Some people will try to convince you buy a manual car today because they still claim it performs and make you feel better for 6-8 hours of driving.

as a software engineer, i feel a real loss of identity right now. for a long time i defined myself in part by the act of writing code. the pride in a hard-earned solution was part of who i was. now i watch AI accomplish in seconds what took me hours. i find myself caught between relief and mourning, awe and anxiety. the craft that shaped me is suddenly eclipsed by a machine. who am i now?

lazygit ten sonra kendim yazdığımı hatırlamıyorum