davideisinger.com

My personal website
Log | Files | Refs | README

simonwillison-net-0ovi9d.txt (13492B)


      1 [1]Simon Willison’s Weblog
      2 
      3 [2]Subscribe
      4 Sponsored by: [3]WorkOS — Make your app Enterprise Ready with SSO, SCIM, RBAC,
      5 and more.
      6 
      7 Vibe coding and agentic engineering are getting closer than I’d like
      8 
      9 6th May 2026
     10 
     11 I recently talked with Joseph Ruscio about AI coding tools for Heavybit’s High
     12 Leverage podcast: [4]Ep. #9, The AI Coding Paradigm Shift with Simon Willison.
     13 Here are some of my highlights, including my disturbing realization that vibe
     14 coding and agentic engineering have started to converge in my own work.
     15 
     16 One thing I really enjoy about podcasts is that they sometimes push me to think
     17 out loud in a way that exposes an idea I’ve not previously been able to put
     18 into words.
     19 
     20 Vibe coding and agentic engineering are starting to overlap [5]#
     21 
     22 A few weeks after vibe coding was first coined I published [6]Not all
     23 AI-assisted programming is vibe coding (but vibe coding rocks), where I firmly
     24 staked out my belief that “vibe coding” is a very different beast from
     25 responsible use of AI to write code, which I’ve since started to call [7]
     26 agentic engineering.
     27 
     28 When Joseph brought up the distinction between the two I had a sudden
     29 realization that they’re not nearly as distinct for me as they used to be:
     30 
     31     Weirdly though, those things have started to blur for me already, which is
     32     quite upsetting.
     33 
     34     I thought we had a very clear delineation where vibe coding is the thing
     35     where you’re not looking at the code at all. You might not even know how to
     36     program. You might be a non-programmer who asks for a thing, and gets a
     37     thing, and if the thing works, then great! And if it doesn’t, you tell it
     38     that it doesn’t work and cross your fingers.
     39 
     40     But at no point are you really caring about the code quality or any of
     41     those additional constraints. And my take on vibe coding was that it’s
     42     fantastic, provided you understand when it can be used and when it can’t.
     43 
     44     A personal tool for you, where if there’s a bug it hurts only you, go
     45     ahead!
     46 
     47     If you’re building software for other people, vibe coding is grossly
     48     irresponsible because it’s other people’s information. Other people get
     49     hurt by your stupid bugs. You need to have a higher level than that.
     50 
     51     This contrasts with agentic engineering where you are a professional
     52     software engineer. You understand security and maintainability and
     53     operations and performance and so forth. You’re using these tools to the
     54     highest of your own ability. I’m finding the scope of challenges I can take
     55     on has gone up by a significant amount because I’ve got the support of
     56     these tools.
     57 
     58     But I’m still leaning on my 25 years of experience as a software engineer.
     59 
     60     The goal is to build high quality production systems: if you’re building
     61     lower quality stuff faster, I think that’s bad. I want to build higher
     62     quality stuff faster. I want everything I’m building to be better in every
     63     way than it was before.
     64 
     65     The problem is that as the coding agents get more reliable, I’m not
     66     reviewing every line of code that they write anymore, even for my
     67     production level stuff.
     68 
     69     I know full well that if you ask Claude Code to build a JSON API endpoint
     70     that runs a SQL query and outputs the results as JSON, it’s just going to
     71     do it right. It’s not going to mess that up. You have it add automated
     72     tests, you have it add documentation, you know it’s going to be good.
     73 
     74     But I’m not reviewing that code. And now I’ve got that feeling of guilt: if
     75     I haven’t reviewed the code, is it really responsible for me to use this in
     76     production?
     77 
     78     The thing that really helps me is thinking back to when I’ve worked at
     79     larger organizations where I’ve been an engineering manager. Other teams
     80     are building software that my team depends on.
     81 
     82     If another team hands over something and says, “hey, this is the image
     83     resize service, here’s how to use it to resize your images”... I’m not
     84     going to go and read every line of code that they wrote.
     85 
     86     I’m going to look at their documentation and I’m going to use it to resize
     87     some images. And then I’m going to start shipping my own features. And if I
     88     start running into problems where the image resizer thing appears to have
     89     bugs or the performance isn’t good, that’s when I might dig into their Git
     90     repositories and see what’s going on. But for the most part I treat that as
     91     a semi-black box that I don’t look at until I need to.
     92 
     93     I’m starting to treat the agents in the same way. And it still feels
     94     uncomfortable, because human beings are accountable for what they do. A
     95     team can build a reputation. I can say “I trust that team over there. They
     96     built good software in the past. They’re not going to build something
     97     rubbish because that affects their professional reputations.”
     98 
     99     Claude Code does not have a professional reputation! It can’t take
    100     accountability for what it’s done. But it’s been proving itself anyway—time
    101     and time again it’s churning out straightforward things and doing them
    102     right in the style that I like.
    103 
    104 There’s an element of [8]the normalization of deviance here—every time a model
    105 turns out to have written the right code without me monitoring it closely
    106 there’s a risk that I’ll trust it at the wrong moment in the future and get
    107 burned.
    108 
    109 The new challenge of evaluating software [9]#
    110 
    111     It used to be if you found a GitHub repository with a hundred commits and a
    112     good readme and automated tests and stuff, you could be pretty sure that
    113     the person writing that had put a lot of care and attention into that
    114     project.
    115 
    116     And now I can knock out a git repository with a hundred commits and a
    117     beautiful readme and comprehensive tests of every line of code in half an
    118     hour! It looks identical to those projects that have had a great deal of
    119     care and attention. Maybe it is as good as them. I don’t know. I can’t tell
    120     from looking at it. Even for my own projects, I can’t tell.
    121 
    122     So I realized what I value more than the quality of the tests and
    123     documentation is that I want somebody to have used the thing. If you’ve got
    124     a vibe coded thing which you have used every day for the past two weeks,
    125     that’s much more valuable to me than something that you’ve just spat out
    126     and hardly even exercised.
    127 
    128 The bottlenecks have shifted [10]#
    129 
    130     If you can go from producing 200 lines of code a day to 2,000 lines of code
    131     a day, what else breaks? The entire software development lifecycle was, it
    132     turns out, designed around the idea that it takes a day to produce a few
    133     hundred lines of code. And now it doesn’t.
    134 
    135     It’s not just the downstream stuff, it’s the upstream stuff as well. I saw 
    136     [11]a great talk by Jenny Wen, who’s the design leader at Anthropic, where
    137     she said we have all of these design processes that are based around the
    138     idea that you need to get the design right—because if you hand it off to
    139     the engineers and they spend three months building the wrong thing, that’s
    140     catastrophic.
    141 
    142     There’s this whole very extensive design process that you put in place
    143     because that design results in expensive work. But if it doesn’t take three
    144     months to build, maybe the design process can be a whole lot riskier
    145     because cost, if you get something wrong, has been reduced so much.
    146 
    147 Why I’m still not afraid for my career [12]#
    148 
    149     When I look at my conversations with the agents, it’s very clear to me that
    150     this is moon language for the vast majority of human beings.
    151 
    152     There are a whole bunch of reasons I’m not scared that my career as a
    153     software engineer is over now that computers can write their own code,
    154     partly because these things are amplifiers of existing experience. If you
    155     know what you’re doing, you can run so much faster with them. [...]
    156 
    157     I’m constantly reminded as I work with these tools how hard the thing that
    158     we do is. Producing software is a ferociously difficult thing to do. And
    159     you could give me all of the AI tools in the world and what we’re trying to
    160     achieve here is still really difficult. [...]
    161 
    162     Matthew Yglesias, who’s a political commentator, yesterday [13]tweeted,
    163     “Five months in, I think I’ve decided that I don’t want to vibecode — I
    164     want professionally managed software companies to use AI coding assistance
    165     to make more/better/cheaper software products that they sell to me for
    166     money.” And that feels about right to me. I can plumb my house if I watch
    167     enough YouTube videos on plumbing. I would rather hire a plumber.
    168 
    169 On the threat to SaaS providers of companies rolling their own solutions
    170 instead:
    171 
    172     I just realized it’s the thing I said earlier about how I only want to use
    173     your side project if you’ve used it for a few weeks. The enterprise version
    174     of that is I don’t want a CRM unless at least two other giant enterprises
    175     have successfully used that CRM for six months. [...] You want solutions
    176     that are proven to work before you take a risk on them.
    177 
    178 Posted [14]6th May 2026 at 2:24 pm · Follow me on [15]Mastodon, [16]Bluesky, 
    179 [17]Twitter or [18]subscribe to my newsletter
    180 
    181 More recent articles
    182 
    183   • [19]Notes on the xAI/Anthropic data center deal - 7th May 2026
    184   • [20]Live blog: Code w/ Claude 2026 - 6th May 2026
    185 
    186 This is Vibe coding and agentic engineering are getting closer than I’d like by
    187 Simon Willison, posted on [21]6th May 2026.
    188 
    189 [22] ai 2016 [23] generative-ai 1785 [24] llms 1751 [25] podcast-appearances 39
    190 [26] vibe-coding 90 [27] coding-agents 201 [28] agentic-engineering 50
    191 
    192 Next: [29]Live blog: Code w/ Claude 2026
    193 
    194 Previous: [30]LLM 0.32a0 is a major backwards-compatible refactor
    195 
    196 Monthly briefing
    197 
    198 Sponsor me for $10/month and get a curated email digest of the month's most
    199 important LLM developments.
    200 
    201 Pay me to send you less!
    202 
    203 [31] Sponsor & subscribe
    204 
    205   • [32]Disclosures
    206   • [33]Colophon
    207   • ©
    208   • [34]2002
    209   • [35]2003
    210   • [36]2004
    211   • [37]2005
    212   • [38]2006
    213   • [39]2007
    214   • [40]2008
    215   • [41]2009
    216   • [42]2010
    217   • [43]2011
    218   • [44]2012
    219   • [45]2013
    220   • [46]2014
    221   • [47]2015
    222   • [48]2016
    223   • [49]2017
    224   • [50]2018
    225   • [51]2019
    226   • [52]2020
    227   • [53]2021
    228   • [54]2022
    229   • [55]2023
    230   • [56]2024
    231   • [57]2025
    232   • [58]2026
    233    234 
    235 
    236 References:
    237 
    238 [1] https://simonwillison.net/
    239 [2] https://simonwillison.net/about/#subscribe
    240 [3] https://fandf.co/4daZLgh
    241 [4] https://www.heavybit.com/library/podcasts/high-leverage/ep-9-the-ai-coding-paradigm-shift-with-simon-willison
    242 [5] https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/#vibe-coding-and-agentic-engineering-are-starting-to-overlap
    243 [6] https://simonwillison.net/2025/Mar/19/vibe-coding/
    244 [7] https://simonwillison.net/guides/agentic-engineering-patterns/what-is-agentic-engineering/
    245 [8] https://simonwillison.net/2025/Dec/10/normalization-of-deviance/
    246 [9] https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/#the-new-challenge-of-evaluating-software
    247 [10] https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/#the-bottlenecks-have-shifted
    248 [11] https://simonwillison.net/2026/Jan/24/dont-trust-the-process/
    249 [12] https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/#why-i-m-still-not-afraid-for-my-career
    250 [13] https://twitter.com/mattyglesias/status/2049105745132585161
    251 [14] https://simonwillison.net/2026/May/6/
    252 [15] https://fedi.simonwillison.net/@simon
    253 [16] https://bsky.app/profile/simonwillison.net
    254 [17] https://twitter.com/simonw
    255 [18] https://simonwillison.net/about/#subscribe
    256 [19] https://simonwillison.net/2026/May/7/xai-anthropic/
    257 [20] https://simonwillison.net/2026/May/6/code-w-claude-2026/
    258 [21] https://simonwillison.net/2026/May/6/
    259 [22] https://simonwillison.net/tags/ai/
    260 [23] https://simonwillison.net/tags/generative-ai/
    261 [24] https://simonwillison.net/tags/llms/
    262 [25] https://simonwillison.net/tags/podcast-appearances/
    263 [26] https://simonwillison.net/tags/vibe-coding/
    264 [27] https://simonwillison.net/tags/coding-agents/
    265 [28] https://simonwillison.net/tags/agentic-engineering/
    266 [29] https://simonwillison.net/2026/May/6/code-w-claude-2026/
    267 [30] https://simonwillison.net/2026/Apr/29/llm/
    268 [31] https://github.com/sponsors/simonw/
    269 [32] https://simonwillison.net/about/#disclosures
    270 [33] https://simonwillison.net/about/#about-site
    271 [34] https://simonwillison.net/2002/
    272 [35] https://simonwillison.net/2003/
    273 [36] https://simonwillison.net/2004/
    274 [37] https://simonwillison.net/2005/
    275 [38] https://simonwillison.net/2006/
    276 [39] https://simonwillison.net/2007/
    277 [40] https://simonwillison.net/2008/
    278 [41] https://simonwillison.net/2009/
    279 [42] https://simonwillison.net/2010/
    280 [43] https://simonwillison.net/2011/
    281 [44] https://simonwillison.net/2012/
    282 [45] https://simonwillison.net/2013/
    283 [46] https://simonwillison.net/2014/
    284 [47] https://simonwillison.net/2015/
    285 [48] https://simonwillison.net/2016/
    286 [49] https://simonwillison.net/2017/
    287 [50] https://simonwillison.net/2018/
    288 [51] https://simonwillison.net/2019/
    289 [52] https://simonwillison.net/2020/
    290 [53] https://simonwillison.net/2021/
    291 [54] https://simonwillison.net/2022/
    292 [55] https://simonwillison.net/2023/
    293 [56] https://simonwillison.net/2024/
    294 [57] https://simonwillison.net/2025/
    295 [58] https://simonwillison.net/2026/