davideisinger.com

My personal website
Log | Files | Refs | README

simonwillison-net-tmayz4.txt (4817B)


      1 [1]Simon Willison’s Weblog
      2 
      3 [2]Subscribe
      4 
      5 [3]Just Talk To It - the no-bs Way of Agentic Engineering. Peter Steinberger's
      6 long, detailed description of his current process for using Codex CLI and GPT-5
      7 Codex. This is information dense and full of actionable tips, plus plenty of
      8 strong opinions about the differences between Claude 4.5 an GPT-5:
      9 
     10     While Claude reacts well to 🚨 SCREAMING ALL-CAPS 🚨 commands that threaten
     11     it that it will imply ultimate failure and 100 kittens will die if it runs
     12     command X, that freaks out GPT-5. (Rightfully so). So drop all of that and
     13     just use words like a human.
     14 
     15 Peter is a heavy user of parallel agents:
     16 
     17     I've completely moved to codex cli as daily driver. I run between 3-8 in
     18     parallel in a 3x3 terminal grid, most of them [4]in the same folder, some
     19     experiments go in separate folders. I experimented with worktrees, PRs but
     20     always revert back to this setup as it gets stuff done the fastest.
     21 
     22 He shares my preference for CLI utilities over MCPs:
     23 
     24     I can just refer to a cli by name. I don't need any explanation in my
     25     agents file. The agent will try $randomcrap on the first call, the cli will
     26     present the help menu, context now has full info how this works and from
     27     now on we good. I don't have to pay a price for any tools, unlike MCPs
     28     which are a constant cost and garbage in my context. Use GitHub's MCP and
     29     see 23k tokens gone. Heck, they did make it better because it was almost
     30     50.000 tokens when it first launched. Or use the gh cli which has basically
     31     the same feature set, models already know how to use it, and pay zero
     32     context tax.
     33 
     34 It's worth reading the [5]section on why he abandoned spec driven development
     35 in full.
     36 
     37 Posted [6]14th October 2025 at 9:26 pm
     38 
     39 Recent articles
     40 
     41   • [7]A new SQL-powered permissions system in Datasette 1.0a20 - 4th November
     42     2025
     43   • [8]New prompt injection papers: Agents Rule of Two and The Attacker Moves
     44     Second - 2nd November 2025
     45   • [9]Hacking the WiFi-enabled color screen GitHub Universe conference badge -
     46     28th October 2025
     47 
     48 [10] ai 1658 [11] generative-ai 1463 [12] llms 1430 [13]
     49 ai-assisted-programming 265 [14] model-context-protocol 22 [15] coding-agents 
     50 88 [16] claude-code 51 [17] codex-cli 10 [18] parallel-agents 6
     51 
     52 Monthly briefing
     53 
     54 Sponsor me for $10/month and get a curated email digest of the month's most
     55 important LLM developments.
     56 
     57 Pay me to send you less!
     58 
     59 [19] Sponsor & subscribe
     60 
     61   • [20]Colophon
     62   • ©
     63   • [21]2002
     64   • [22]2003
     65   • [23]2004
     66   • [24]2005
     67   • [25]2006
     68   • [26]2007
     69   • [27]2008
     70   • [28]2009
     71   • [29]2010
     72   • [30]2011
     73   • [31]2012
     74   • [32]2013
     75   • [33]2014
     76   • [34]2015
     77   • [35]2016
     78   • [36]2017
     79   • [37]2018
     80   • [38]2019
     81   • [39]2020
     82   • [40]2021
     83   • [41]2022
     84   • [42]2023
     85   • [43]2024
     86   • [44]2025
     87 
     88 
     89 References:
     90 
     91 [1] https://simonwillison.net/
     92 [2] https://simonwillison.net/about/#subscribe
     93 [3] https://steipete.me/posts/just-talk-to-it
     94 [4] https://x.com/steipete/status/1977771686176174352
     95 [5] https://steipete.me/posts/just-talk-to-it#do-you-do-spec-driven-development
     96 [6] https://simonwillison.net/2025/Oct/14/
     97 [7] https://simonwillison.net/2025/Nov/4/datasette-10a20/
     98 [8] https://simonwillison.net/2025/Nov/2/new-prompt-injection-papers/
     99 [9] https://simonwillison.net/2025/Oct/28/github-universe-badge/
    100 [10] https://simonwillison.net/tags/ai/
    101 [11] https://simonwillison.net/tags/generative-ai/
    102 [12] https://simonwillison.net/tags/llms/
    103 [13] https://simonwillison.net/tags/ai-assisted-programming/
    104 [14] https://simonwillison.net/tags/model-context-protocol/
    105 [15] https://simonwillison.net/tags/coding-agents/
    106 [16] https://simonwillison.net/tags/claude-code/
    107 [17] https://simonwillison.net/tags/codex-cli/
    108 [18] https://simonwillison.net/tags/parallel-agents/
    109 [19] https://github.com/sponsors/simonw/
    110 [20] https://simonwillison.net/about/#about-site
    111 [21] https://simonwillison.net/2002/
    112 [22] https://simonwillison.net/2003/
    113 [23] https://simonwillison.net/2004/
    114 [24] https://simonwillison.net/2005/
    115 [25] https://simonwillison.net/2006/
    116 [26] https://simonwillison.net/2007/
    117 [27] https://simonwillison.net/2008/
    118 [28] https://simonwillison.net/2009/
    119 [29] https://simonwillison.net/2010/
    120 [30] https://simonwillison.net/2011/
    121 [31] https://simonwillison.net/2012/
    122 [32] https://simonwillison.net/2013/
    123 [33] https://simonwillison.net/2014/
    124 [34] https://simonwillison.net/2015/
    125 [35] https://simonwillison.net/2016/
    126 [36] https://simonwillison.net/2017/
    127 [37] https://simonwillison.net/2018/
    128 [38] https://simonwillison.net/2019/
    129 [39] https://simonwillison.net/2020/
    130 [40] https://simonwillison.net/2021/
    131 [41] https://simonwillison.net/2022/
    132 [42] https://simonwillison.net/2023/
    133 [43] https://simonwillison.net/2024/
    134 [44] https://simonwillison.net/2025/