<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Symphonious</title>
    <link>https://www.symphonious.net/</link>
    <description>Recent content on Symphonious</description>
    <image>
      <url>https://www.symphonious.net/images/after-hours-icon.png</url>
      <title>Symphonious</title>
      <link>https://www.symphonious.net/</link>
      <width>144</width>
      <height>144</height>
    </image>
    <generator>Hugo -- gohugo.io</generator>
    <language>en-AU</language>
    <copyright>Adrian Sutton 2004 -</copyright>
    <lastBuildDate>Mon, 02 Nov 2026 09:20:49 +1000</lastBuildDate><atom:link href="https://www.symphonious.net/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>New Blog Design</title>
      <link>https://www.symphonious.net/2026/11/02/new-blog-design/</link>
      <pubDate>Mon, 02 Nov 2026 09:20:49 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/11/02/new-blog-design/</guid>
      <description>&lt;p&gt;It&amp;rsquo;s been around 10 to 15 years since the last time this blog&amp;rsquo;s design changed
so it seemed like it was due. The cover image is from a gorgeous spot on Lake
Moogerah which has the most amazing sunsets. I&amp;rsquo;ve been fortunate enough to
travel out there for a few Church small group retreats and it&amp;rsquo;s always very
special. Such a wonderful place to slow down, spend time with God and appreciate
the beauty of his creation.&lt;/p&gt;
&lt;p&gt;This is also the first time I&amp;rsquo;ve redesigned the site since AI got good and wow
it made it a lot easier. It&amp;rsquo;s a relatively bland design with the cover image
providing pretty much all the personality but that&amp;rsquo;s true of every other design
this site has had too. It&amp;rsquo;s a content focussed site, not a showy design one and
I am definitely not a designer. Even so, being able to have AI pull out matching
colours from my photo, deal with all the sizing details, write the actual Hugo
theme and pull it all together made it a snap. In fact, I was just instructing
an OpenClaw agent from my phone via Slack and it sent back image mock ups to
start and then published a private version of the site for me to test.&lt;/p&gt;
&lt;p&gt;Beyond my website, more and more I find I&amp;rsquo;m getting good results from just
telling AI to &amp;ldquo;make it pretty&amp;rdquo;. Great designers will always have a job, but
there&amp;rsquo;s an awful lot of stuff that previously got no styling or design at all
that is now way nicer because AI makes it easy. It may just be lipstick on a
pig, but the pig still looks a lot nicer than it did.&lt;/p&gt;
&lt;p&gt;Oh and it implemented search which was lost when I moved to a statically
generated site.&lt;/p&gt;
&lt;p&gt;P.S: The link to &lt;a href=&#34;https://slots.symphonious.net/&#34;&gt;slot finder&lt;/a&gt; is now in the
footer. Hopefully this doesn&amp;rsquo;t cause delays for the next Ethereum hard fork.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Whiteboarding a Spec</title>
      <link>https://www.symphonious.net/2026/10/31/whiteboarding-a-spec/</link>
      <pubDate>Sat, 31 Oct 2026 18:35:18 -0400</pubDate>
      <guid>https://www.symphonious.net/2026/10/31/whiteboarding-a-spec/</guid>
      <description>&lt;p&gt;I&amp;rsquo;ve been in New York the past couple of weeks planning out a new project. Part
of that work has been working out &lt;em&gt;how&lt;/em&gt; the new thing will be built - what
protocol and contract changes are needed and how does it all fit in with the
existing system? Ultimately we want to create a specification that we have a
reasonably high confidence in, but getting that means working through a lot of
scenarios and potential designs which is best done on a whiteboard. We found a
really powerful way of using AI to help without winding up with a giant pile of
AI slop.&lt;/p&gt;&lt;figure class=&#34;feature-image&#34; data-pagefind-ignore&gt;
  &lt;img src=&#34;https://www.symphonious.net/images/posts/whiteboarding-a-spec.webp&#34; srcset=&#34;https://www.symphonious.net/images/posts/whiteboarding-a-spec_hu_c1432b92bd4dad0c.webp 480w, https://www.symphonious.net/images/posts/whiteboarding-a-spec_hu_e2613579dc82eebf.webp 740w, https://www.symphonious.net/images/posts/whiteboarding-a-spec_hu_ca5532145c8ddcd.webp 1100w, https://www.symphonious.net/images/posts/whiteboarding-a-spec.webp 1480w&#34; sizes=&#34;(max-width: 600px) calc(100vw - 40px), (max-width: 1000px) calc(100vw - 96px), 740px&#34; width=&#34;1480&#34; height=&#34;740&#34; alt=&#34;Whiteboard sketches becoming a structured specification, with a hand making pencil corrections.&#34; decoding=&#34;async&#34;&gt;&lt;figcaption&gt;The thinking happens in the room. AI helps get it onto the page.&lt;/figcaption&gt;
&lt;/figure&gt;
&lt;p&gt;Prior to getting together in person, there was a good idea of the customer
requirements and a steel-man approach we could use. That was pretty much
entirely AI slop, but AI slop is a great way to play around with a PoC and try
out many different ideas quickly. From there we shut ourselves in a conference
room and did what engineers have been doing forever - we talked through the
design, scribbled on a whiteboard and tried to find ways to either improve or
break it. What&amp;rsquo;s relatively new for me (been doing it for most of the year) is
to also dial into a Google Meet and have it record a transcription of everything
in the room. Prior to AI you&amp;rsquo;d never read such a transcription - it might be
useful to search and find specific decisions but mostly it went unused. With AI,
each evening I got it to summarise the transcript and update a notion page that
captured our work. More likely to be useful at some point - especially after
kicking the AI a few times to make it an up to date record, not a historical
summary of all discussions. That page became a pretty good summary of decisions
we&amp;rsquo;d made and where we were up to - still clearly AI slop but usable.&lt;/p&gt;
&lt;p&gt;What really surprised me though, was that after a few days of going back and
forward on design options, changing our minds, breaking things, fixing them,
getting confused ourselves and referencing older designs we had to change, we
just ask AI to read the transcripts and write a spec. It did have the benefit of
starting from our existing specs repo which gave a large corpus of examples for
the style to use. It churned out a quite readable and 90% accurate spec. We
still had to go through it all and made quite a lot of corrections but it saved
us a day or two of manual writing, and even without AI there&amp;rsquo;s always a ton of
things to fix.&lt;/p&gt;
&lt;p&gt;I&amp;rsquo;ve now read the spec end to end a few times as well as digging into specific
parts to help build task break downs and things like that - it reads really well
and while there&amp;rsquo;s a few AI-isms through it, it&amp;rsquo;s definitely not slop and is
really useful. Each milestone in the project still has a task to &amp;ldquo;write&amp;rdquo; the
relevant piece of the spec to make sure it is updated with any final adjustments
we make from actual implementation and to remove the last few AI-isms but it was
a significant time saver.&lt;/p&gt;
&lt;p&gt;Perhaps even more than the work hours we saved though, is the immediacy of
having spec documents ready. I was only available in-person for two weeks and we
had a lot to get through - normally we&amp;rsquo;d have just made some notes or recordings
and left writing the actual spec for afterwards since it can be done more async.
But then we&amp;rsquo;d be left to review it async rather than being able to do that in
person. Given our first pass review identified a few sections we needed to
significantly rework because we&amp;rsquo;d missed something, the AI generated spec let us
resolve that in person with high bandwidth communication. Much better than
having to schedule a lot of early morning calls for me.&lt;/p&gt;
&lt;p&gt;And of course, afterwards we can have AI generate really bad songs about
everything we discussed, full of all kinds of in jokes and weird sayings that
happened in the room. But it only really makes sense if you were in the room
where it happened&amp;hellip;&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Just Vibe Code It Yourself</title>
      <link>https://www.symphonious.net/2026/08/16/just-vibe-code-it-yourself/</link>
      <pubDate>Sun, 16 Aug 2026 19:22:52 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/08/16/just-vibe-code-it-yourself/</guid>
      <description>&lt;p&gt;When I started Moolah, I built it as a vibe coded, but &amp;ldquo;serious&amp;rdquo; project. I
spent time planning features, thinking through design and setting up systems to
make sure it was high quality. That takes a fair bit of time and effort, so
while it comes together faster than if I wrote it by hand, it still &lt;em&gt;felt&lt;/em&gt; like
a software project. For many things, that level of engineering isn&amp;rsquo;t required
and you can truly build something that is just for yourself. By other people&amp;rsquo;s
standards it would be AI slop, but because it is perfectly tailored to your
needs, for you it&amp;rsquo;s great. That presents quite a challenge for the wide range of
small development shops building high quality, but relatively small software.
More and more, people really can just build it for themselves and get exactly
what they want. It won&amp;rsquo;t &lt;em&gt;out sell&lt;/em&gt; your product, but if enough people start
doing it, you still wind up without any customers.&lt;/p&gt;&lt;figure class=&#34;feature-image&#34; data-pagefind-ignore&gt;
  &lt;img src=&#34;https://www.symphonious.net/images/posts/just-vibe-code-it-yourself.webp&#34; srcset=&#34;https://www.symphonious.net/images/posts/just-vibe-code-it-yourself_hu_b3b63cc73e1852cb.webp 480w, https://www.symphonious.net/images/posts/just-vibe-code-it-yourself_hu_296727de76c5d789.webp 740w, https://www.symphonious.net/images/posts/just-vibe-code-it-yourself.webp 884w&#34; sizes=&#34;(max-width: 600px) calc(100vw - 40px), (max-width: 1000px) calc(100vw - 96px), 740px&#34; width=&#34;884&#34; height=&#34;442&#34; alt=&#34;Hands assembling a wooden laptop with a recipe, calendar and checklist, tailoring software to personal needs.&#34; decoding=&#34;async&#34;&gt;
&lt;/figure&gt;
&lt;p&gt;There&amp;rsquo;s plenty of software that&amp;rsquo;s still safe from this. Vibe-coding Microsoft
Word or Excel would be pretty crazy. The basics generally work extremely well
for people and they have a large depth of functionality that usually scratches
whatever itch people typically have. Plus compatibility is a very big deal - a
vibe coded word processor that either doesn&amp;rsquo;t support Word documents or renders
them slightly differently is not generally useful. Free software also has an
advantage because if it does what you need for free, why bother rewriting it?&lt;/p&gt;
&lt;p&gt;The software I can see being most affected are the more niche products, often by
indie developers. For example I use Paprika for tracking recipes - it&amp;rsquo;s very
good but has a few things that I wish worked slightly differently. On iOS it&amp;rsquo;s
priced pretty reasonably so I bought it there. The problem is, you have to buy
each platform separately and the Mac prices are way higher than I&amp;rsquo;m willing to
pay for it. That&amp;rsquo;s not a complaint really, the business model seems to work for
them and I hope it keeps them in business. However, it makes me wonder if I
could just vibe code a web based solution that, while probably way worse than
Paprika overall, would work perfectly for me. The main thing that has made me
hold off is that Paprika is very good at scraping recipes from any website and
then being able to display it in a standard format - much easier for cooking
with. Building that scraping code itself is hard, but shelling out to
&lt;code&gt;claude -p&lt;/code&gt; would probably do the trick just fine. Beyond that, recipe
management is a pretty standard CRUD database app.&lt;/p&gt;
&lt;p&gt;The thing is, a &lt;em&gt;lot&lt;/em&gt; of software is a pretty standard CRUD database. At LMAX
maybe 5% of the code was really smart, high performance, exchange specific code.
The other 95% was just loading the resulting data, transforming it and spitting
out a report. Moolah is just a database with a highly customised UI. My wife&amp;rsquo;s
colleague vibe coded a system for tracking all the music student&amp;rsquo;s private
lessons and the tutor&amp;rsquo;s time sheeting.&lt;/p&gt;
&lt;p&gt;A lot of software that looks &amp;ldquo;big&amp;rdquo; and complex to replace is actually just a lot
of smaller, simple systems hooked up together. And often augmenting that central
system are hundreds of spreadsheets that are the inputs and outputs and that get
tweaked to scratch the itches people have. Even those big systems are at risk of
gradually being eaten by smaller, more personally tailored systems as people
find they can solve their needs better with something they vibe coded. And they
can generally still integrate with the bigger system because the UI was really
all those spreadsheets anyway and they&amp;rsquo;re easy to generate.&lt;/p&gt;
&lt;p&gt;There&amp;rsquo;s going to be a ton of security, reliability and accuracy problems with
these new systems, but the reliability and accuracy are likely to get ironed out
pretty quickly because the developers &lt;em&gt;are&lt;/em&gt; the users and they&amp;rsquo;re the experts in
their field. It&amp;rsquo;s the ultimate form of XP&amp;rsquo;s on-site customer. Security will be
much harder to solve, but given how poorly professional software is holding up
to AI analysis these days, maybe the answer to that is just good AI tooling and
analysis, plus a few really solid foundational libraries.&lt;/p&gt;
&lt;p&gt;It&amp;rsquo;s probably not the end of software development as a career, but it will
certainly take a large chunk out of it. That&amp;rsquo;s a little scary, but as someone
who go into programming as a hobby and then turned that hobby into a career, I&amp;rsquo;m
kind of excited to see the true power of computing be opened up to a much
broader range of people.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>My New Best Friend Is ChatGPT</title>
      <link>https://www.symphonious.net/2026/07/07/my-new-best-friend-is-chatgpt/</link>
      <pubDate>Tue, 07 Jul 2026 15:49:13 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/07/07/my-new-best-friend-is-chatgpt/</guid>
      <description>&lt;p&gt;I&amp;rsquo;ve switched from Claude to ChatGPT as my main AI assistant. For work I&amp;rsquo;ve been
toggling between the two but I only need to pay for one subscription personally.
The current Claude subscription just ran out so now my new best friend is
ChatGPT&amp;hellip;&lt;/p&gt;
&lt;p style=&#34;text-align: center&#34;&gt;&lt;img src=&#34;https://www.symphonious.net/images/posts/2026-07-07-my-new-best-friend-is-chatgpt/new-best-friend-chatgpt.png&#34; alt=&#34;Meme graphic with the text My New Best Friend Is ChatGPT&#34; /&gt;&lt;/p&gt;
&lt;p&gt;The biggest issue with Claude is that they just don&amp;rsquo;t seem to have enough
capacity for what they&amp;rsquo;re selling. Response times have become quite long and are
dominated by the time before any tokens actually get spent, clearly just waiting
to be allocated some GPU time. Plus I&amp;rsquo;ve regularly had problems where they rate
limit all responses even though I&amp;rsquo;m well within my paid for quota. It&amp;rsquo;s annoying
enough that AI plans have completely opaque, and unquantified token allowances,
but to not actually be able to use what you pay for is ridiculous. I don&amp;rsquo;t
expect ChatGPT will be perfect, but its certainly a lot faster from my early
experience.&lt;/p&gt;
&lt;p&gt;The Claude Mac app is also terrible, but I&amp;rsquo;ve heard very good things about the
ChatGPT app, and again early indications are promising. I may still use the
codex CLI tool, but the Codex app feels like its at least worth a try. The
ChatGPT app already feels far richer and more responsive.&lt;/p&gt;
&lt;p&gt;I&amp;rsquo;ve also started experimenting with &lt;code&gt;omp&lt;/code&gt; as a harness (thanks for the tip
Rodrigo!). I&amp;rsquo;m not sure if it&amp;rsquo;s particularly better or worse at the moment but I
like the idea of using a third party harness so I can more easily switch models
without having to switch UIs and convert from CLAUDE.md to AGENTS.md etc. It&amp;rsquo;s
slightly surprising that I haven&amp;rsquo;t found working in traditional IDEs with their
AI integrations to work particularly well. That may have improved since I last
tried it, and I certainly miss having a really great IDE, but somehow the many
parallel AI agents doesn&amp;rsquo;t fit well with a traditional IDE. There&amp;rsquo;s very likely
a real opportunity for however can really crack the right way to build a great
AI-first IDE. Plenty of people are working on it and some look like they have
potential but I haven&amp;rsquo;t seen anything that really feels right to me yet.&lt;/p&gt;
&lt;p&gt;So many levels of experimentation in this new world&amp;hellip; Still fun!&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Ads Evading Pi-hole with iOS 27</title>
      <link>https://www.symphonious.net/2026/06/18/ads-evading-pi-hole-with-ios-27/</link>
      <pubDate>Thu, 18 Jun 2026 12:48:00 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/06/18/ads-evading-pi-hole-with-ios-27/</guid>
      <description>&lt;p&gt;I YOLO installed the iOS 27 beta and noticed ads creeping back into apps that
have been clean for years. The &lt;a href=&#34;https://pi-hole.net/&#34;&gt;Pi-hole&lt;/a&gt; was still running
and still showing blocked queries — the ads were just loading anyway.&lt;/p&gt;&lt;figure class=&#34;feature-image&#34; data-pagefind-ignore&gt;
  &lt;img src=&#34;https://www.symphonious.net/images/posts/ads-evading-pi-hole.webp&#34; srcset=&#34;https://www.symphonious.net/images/posts/ads-evading-pi-hole_hu_a9a68e9815b37cf9.webp 480w, https://www.symphonious.net/images/posts/ads-evading-pi-hole_hu_1e413f5e34d39b70.webp 740w, https://www.symphonious.net/images/posts/ads-evading-pi-hole.webp 884w&#34; sizes=&#34;(max-width: 600px) calc(100vw - 40px), (max-width: 1000px) calc(100vw - 96px), 740px&#34; width=&#34;884&#34; height=&#34;442&#34; alt=&#34;Advertising bypassing a home network filter over a cellular connection to a smartphone.&#34; decoding=&#34;async&#34;&gt;
&lt;/figure&gt;
&lt;p&gt;The culprit was &amp;ldquo;Connectivity Assist&amp;rdquo; in the Wi-Fi settings, aka Cellular
Fallback. When the Pi-hole blackholes an ad domain the request fails, iOS
decides the Wi-Fi must be flaky, and helpfully retries the same request over
mobile data — where there&amp;rsquo;s no Pi-hole in the path. So the ad not only loads, it
burns through my data allowance to do it.&lt;/p&gt;
&lt;p&gt;Flipped the switch off and ads are properly being blocked again.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Moolah Diaries - A Month Later</title>
      <link>https://www.symphonious.net/2026/06/01/moolah-diaries-a-month-later/</link>
      <pubDate>Mon, 01 Jun 2026 18:16:13 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/06/01/moolah-diaries-a-month-later/</guid>
      <description>&lt;p&gt;It&amp;rsquo;s been a month since
&lt;a href=&#34;https://www.symphonious.net/2026/04/28/moolah-diaries-going-all-in/&#34;&gt;I went all in on using Moolah&lt;/a&gt; so
time to check in on how its going.&lt;/p&gt;&lt;figure class=&#34;feature-image&#34; data-pagefind-ignore&gt;
  &lt;img src=&#34;https://www.symphonious.net/images/posts/moolah-a-month-later.webp&#34; srcset=&#34;https://www.symphonious.net/images/posts/moolah-a-month-later_hu_d3ccdf0266bd0e77.webp 480w, https://www.symphonious.net/images/posts/moolah-a-month-later_hu_53232834affe318c.webp 740w, https://www.symphonious.net/images/posts/moolah-a-month-later.webp 884w&#34; sizes=&#34;(max-width: 600px) calc(100vw - 40px), (max-width: 1000px) calc(100vw - 96px), 740px&#34; width=&#34;884&#34; height=&#34;442&#34; alt=&#34;A personal finance dashboard beside a ledger, calendar and coins, with unfinished components waiting nearby.&#34; decoding=&#34;async&#34;&gt;
&lt;/figure&gt;
&lt;p&gt;Firstly the bugs - there are still some but generally in the functions I&amp;rsquo;m
actually using there aren&amp;rsquo;t many and things are generally pretty polished. Most
importantly regressions are rare so as I work through fixing things it keeps
getting consistently better. Most of the issues now are more of the UI polish
kind of problem. I&amp;rsquo;m not a UI designer and the AI is &lt;em&gt;definitely&lt;/em&gt; not a UI
designer, so getting any level of polish is pretty challenging. Again, the flows
I actually use and quite smooth and work well for me. That&amp;rsquo;s pretty much inline
with how the web based version of Moolah worked too - quite nice for me, but
very much tailored to me and a bit obscure to everyone else.&lt;/p&gt;
&lt;p&gt;What is interesting is the amount of functionality that just gets added but not
really used. It&amp;rsquo;s so easy to come up with an idea and let AI loose implementing
it that I have tended to do that. Actually getting it to the point where it
really works and I&amp;rsquo;d actually trust it requires more attention though. Net
result is there are a bunch of features that are promising and would make my
life easier, but haven&amp;rsquo;t actually been finished off enough to be used. In fact
every so often I throw some idea at the AI and it just points out it is already
implemented but not actually hooked up.&lt;/p&gt;
&lt;p&gt;There are some definite wins though - Moolah native has multi-currency support
that Moolah web never did and it extends to stocks and crypto. So far I&amp;rsquo;m only
using the currencies and stocks support but since I&amp;rsquo;m paid in USD its nice to be
able to track the actual USD amount that is received and then the trade to
convert it to AUD accurately. Similarly, stocks are tracked and automatically
pull daily prices rather than needing me to grab a snapshot of the total value
each day so I have a historical record. Crypto support is there and I think good
enough to actually use but having multiple wallets across different chains made
it a bit complex to track prior to the account groups functionality that landed
recently. The biggest barrier however is how to manage the switch over from one
tracking mechanism to the new one. With 10+ years of data it can be a
challenge&amp;hellip;&lt;/p&gt;
&lt;p&gt;Overall it&amp;rsquo;s been a pretty huge success and I still haven&amp;rsquo;t read any of the
code. For anything that is intended for an audience of one, vibe coding is a
perfectly good way to approach it. You need to be prepared to iterate on it with
the AI a bit - it&amp;rsquo;s definitely not just one shot - but you really can build
something that can really solve your needs and be nice for you to use. Even
without programming experience an advanced user could really use AI to build
applications - and AI is &lt;em&gt;way&lt;/em&gt; better at building web apps than native so
there&amp;rsquo;s even more potential there. That&amp;rsquo;s exciting - it&amp;rsquo;s a return to the more
manipulatable, content-creation focussed computing of my youth. In more modern
terms content creation has been very possible and a big focus, but its been more
on creative output (writing, film, music etc) than on creating computer programs
themselves.&lt;/p&gt;
&lt;p&gt;Does this replace &amp;ldquo;real&amp;rdquo; software development? Could you create software that is
good for other people to use? Not with AI by itself but if you know the domain
and have at least a decent sense of UI design, then probably. There are
certainly apps being created by non-programmers that are seeing good adoption
and that doesn&amp;rsquo;t surprise me. I&amp;rsquo;m certainly not fearing for my job, and I don&amp;rsquo;t
think most programmers should, but if you&amp;rsquo;re an average to below average
developer just churning out relatively simple stuff you&amp;rsquo;re in trouble. It&amp;rsquo;s
possible that new developers will find it very difficult to build up the
experience they need to get into the industry, but it&amp;rsquo;s also possible that AI
will provide a new pathway to convert power users into professional programmers.
A large number of our existing programmers actually got started that way -
developing stuff for fun and using that experience to bootstrap careers.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>A REST API for Ethereum Execution Clients</title>
      <link>https://www.symphonious.net/2026/05/30/a-rest-api-for-ethereum-execution-clients/</link>
      <pubDate>Sat, 30 May 2026 16:30:00 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/05/30/a-rest-api-for-ethereum-execution-clients/</guid>
      <description>&lt;p&gt;The JSON-RPC API that Ethereum execution clients expose is genuinely awful to
work with. Every call is a &lt;code&gt;POST&lt;/code&gt; with a &lt;code&gt;jsonrpc&lt;/code&gt; envelope, a &lt;code&gt;method&lt;/code&gt; string
and a positional &lt;code&gt;params&lt;/code&gt; array. Numbers come back as hex strings — block
numbers, balances, gas, timestamps, all of it — so even after you&amp;rsquo;ve wrestled
the request together you&amp;rsquo;re piping the response through something to turn
&lt;code&gt;0x1bc16d674ec80000&lt;/code&gt; back into a number a human can read. It does just about
everything it can to be hostile to a quick &lt;code&gt;curl&lt;/code&gt;.&lt;/p&gt;&lt;figure class=&#34;feature-image&#34; data-pagefind-ignore&gt;
  &lt;img src=&#34;https://www.symphonious.net/images/posts/ethereum-rest-api.webp&#34; srcset=&#34;https://www.symphonious.net/images/posts/ethereum-rest-api_hu_6d165f1063a152e5.webp 480w, https://www.symphonious.net/images/posts/ethereum-rest-api_hu_32c776d17bfc190a.webp 740w, https://www.symphonious.net/images/posts/ethereum-rest-api.webp 884w&#34; sizes=&#34;(max-width: 600px) calc(100vw - 40px), (max-width: 1000px) calc(100vw - 96px), 740px&#34; width=&#34;884&#34; height=&#34;442&#34; alt=&#34;A mechanical adapter turning tangled Ethereum connections into neatly organised resource cards.&#34; decoding=&#34;async&#34;&gt;
&lt;/figure&gt;
&lt;p&gt;Tools like &lt;a href=&#34;https://book.getfoundry.sh/cast/&#34;&gt;Foundry&amp;rsquo;s &lt;code&gt;cast&lt;/code&gt;&lt;/a&gt; make it usable,
and I reach for them constantly, but they&amp;rsquo;re still fiddly for ad-hoc poking and
they don&amp;rsquo;t help at all when something other than your terminal wants to talk to
the client. What really bugs me is that we designed such a bad API in the first
place and then built a layer of tooling to paper over it.&lt;/p&gt;
&lt;p&gt;So, naturally, I built another layer to paper over it.
&lt;a href=&#34;https://github.com/ajsutton/exec-rest-api&#34;&gt;exec-rest-api&lt;/a&gt; is a proxy that sits
in front of any execution client and exposes a sensible REST API, translating
your requests into the JSON-RPC calls the client actually wants.&lt;/p&gt;
&lt;h2 id=&#34;the-awful-api&#34;&gt;The Awful API&lt;/h2&gt;
&lt;p&gt;Here&amp;rsquo;s what asking for the current chain ID looks like today:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;curl -X POST -H &lt;span style=&#34;color:#e6db74&#34;&gt;&amp;#34;Content-Type: application/json&amp;#34;&lt;/span&gt; &lt;span style=&#34;color:#ae81ff&#34;&gt;\
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  --data &lt;span style=&#34;color:#e6db74&#34;&gt;&amp;#39;{&amp;#34;jsonrpc&amp;#34;:&amp;#34;2.0&amp;#34;,&amp;#34;method&amp;#34;:&amp;#34;eth_chainId&amp;#34;,&amp;#34;params&amp;#34;:[],&amp;#34;id&amp;#34;:1}&amp;#39;&lt;/span&gt; &lt;span style=&#34;color:#ae81ff&#34;&gt;\
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  http://localhost:8545
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#75715e&#34;&gt;# {&amp;#34;jsonrpc&amp;#34;:&amp;#34;2.0&amp;#34;,&amp;#34;id&amp;#34;:1,&amp;#34;result&amp;#34;:&amp;#34;0x1&amp;#34;}&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;A &lt;code&gt;POST&lt;/code&gt; with a hand-written envelope, to ask a question that has no parameters,
and the answer is &lt;code&gt;0x1&lt;/code&gt; rather than &lt;code&gt;1&lt;/code&gt;. Want the block number too? That&amp;rsquo;s a
second round of envelope-writing and another hex value to decode. None of this
is hard, exactly, it&amp;rsquo;s just tedious in a way that adds up every single time.&lt;/p&gt;
&lt;p&gt;The same thing through the proxy:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;curl http://127.0.0.1:8080/chain
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#75715e&#34;&gt;# {&amp;#34;chainId&amp;#34;: 1, &amp;#34;networkId&amp;#34;: &amp;#34;1&amp;#34;, &amp;#34;client&amp;#34;: &amp;#34;Geth/v1.13.5...&amp;#34;, ...}&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;A plain &lt;code&gt;GET&lt;/code&gt;, no envelope, decimal numbers, and a bit of extra context thrown
in for free. That&amp;rsquo;s the whole idea.&lt;/p&gt;
&lt;h2 id=&#34;what-it-does&#34;&gt;What It Does&lt;/h2&gt;
&lt;p&gt;The API is organised the way you&amp;rsquo;d actually think about the chain rather than
the way the RPC methods are named. There are resources for &lt;code&gt;/chain&lt;/code&gt;, &lt;code&gt;/blocks&lt;/code&gt;,
&lt;code&gt;/accounts&lt;/code&gt;, &lt;code&gt;/transactions&lt;/code&gt;, &lt;code&gt;/logs&lt;/code&gt;, &lt;code&gt;/traces&lt;/code&gt; and &lt;code&gt;/gas&lt;/code&gt;, plus a
&lt;code&gt;/utils/keccak256&lt;/code&gt; helper for the inevitable moment you need to hash something.
Behind each of those it&amp;rsquo;s still making the &lt;code&gt;eth_getBlockByNumber&lt;/code&gt; and
&lt;code&gt;eth_getBalance&lt;/code&gt; and friends that the client understands — you just don&amp;rsquo;t have
to think about it.&lt;/p&gt;
&lt;p&gt;Hex quantity notation is stripped out everywhere, so numbers are numbers. Errors
come back as &lt;a href=&#34;https://www.rfc-editor.org/rfc/rfc9457&#34;&gt;RFC 9457&lt;/a&gt; problem details
instead of the JSON-RPC error shape, and anything that returns a list uses
&lt;a href=&#34;https://www.rfc-editor.org/rfc/rfc8288&#34;&gt;RFC 8288&lt;/a&gt; cursor pagination rather than
making you guess at ranges. If you do need the raw bytes — a block or
transaction as RLP — content negotiation will give you that instead.&lt;/p&gt;
&lt;p&gt;There&amp;rsquo;s a &lt;code&gt;/health&lt;/code&gt; resource that&amp;rsquo;s genuinely useful for orchestration:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;curl http://127.0.0.1:8080/health/ready
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;&lt;span style=&#34;color:#75715e&#34;&gt;# {&amp;#34;ready&amp;#34;: true, &amp;#34;upstreamReachable&amp;#34;: true, &amp;#34;syncing&amp;#34;: false, ...}&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;That single call rolls up &amp;ldquo;is the upstream reachable&amp;rdquo; and &amp;ldquo;is it still syncing&amp;rdquo;
into one readiness check, which is exactly the thing you otherwise end up
scripting by hand around &lt;code&gt;eth_syncing&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;For the cases where polling is the wrong model, there are Server-Sent Events
streams at &lt;code&gt;/streams/blocks&lt;/code&gt;, &lt;code&gt;/streams/logs&lt;/code&gt;, &lt;code&gt;/streams/pending-transactions&lt;/code&gt;
and &lt;code&gt;/streams/sync-status&lt;/code&gt;. SSE is a much friendlier thing to consume from a
script or a small service than the WebSocket subscription dance, and it works
with &lt;code&gt;curl&lt;/code&gt; directly. There&amp;rsquo;s also a Prometheus &lt;code&gt;/metrics&lt;/code&gt; endpoint and
request-tracing headers — &lt;code&gt;X-Request-ID&lt;/code&gt;, &lt;code&gt;X-Upstream-Method&lt;/code&gt;, &lt;code&gt;X-Block-Height&lt;/code&gt;
— so when something looks wrong you can see exactly which JSON-RPC call your
request turned into.&lt;/p&gt;
&lt;h2 id=&#34;running-it&#34;&gt;Running It&lt;/h2&gt;
&lt;p&gt;The thing I cared most about was that it be trivial to start, because the whole
point is ad-hoc use. If you&amp;rsquo;ve got &lt;code&gt;pipx&lt;/code&gt;:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;pipx install exec-rest-api
&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;exec-rest-api --upstream-http http://localhost:8545
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;There&amp;rsquo;s a plain &lt;code&gt;pip install&lt;/code&gt; too, or a single-file &lt;code&gt;.pyz&lt;/code&gt; you can download from
the releases and run directly:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;./exec-rest-api.pyz --upstream-http http://localhost:8545
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Or Docker, if you&amp;rsquo;d rather not install anything:&lt;/p&gt;
&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;&#34;&gt;&lt;code class=&#34;language-bash&#34; data-lang=&#34;bash&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;docker run --rm -p 8080:8080 &lt;span style=&#34;color:#ae81ff&#34;&gt;\
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  ghcr.io/ajsutton/exec-rest-api:latest &lt;span style=&#34;color:#ae81ff&#34;&gt;\
&lt;/span&gt;&lt;/span&gt;&lt;/span&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;  --upstream-http http://host.docker.internal:8545
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;It works with any execution client — it&amp;rsquo;s only ever speaking standard JSON-RPC
upstream — so point it at Geth, Nethermind, Besu, Reth, whatever you&amp;rsquo;re running.
The same low ceremony means it&amp;rsquo;s just as happy left running as a long-lived
proxy in front of a node as it is spun up for thirty seconds to answer one
question.&lt;/p&gt;
&lt;p&gt;Is this likely to get any real use? Probably not. The current JSON-RPC is really
very entrenched at this point but I do like how easy it is so solve and at least
my home node can now expose a nice API. You&amp;rsquo;ll also note that it differs from
the Ethereum consensus client REST API in some conventions because if I&amp;rsquo;m being
overly opinionated about APIs I may as well run with it and do it the way I
think the Consensus API should have been done. It didn&amp;rsquo;t do something as awful
as use hex for numbers, but putting version numbers in URLs is almost as bad&amp;hellip;&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Moolah Diaries - Going All In</title>
      <link>https://www.symphonious.net/2026/04/28/moolah-diaries-going-all-in/</link>
      <pubDate>Tue, 28 Apr 2026 20:10:26 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/04/28/moolah-diaries-going-all-in/</guid>
      <description>&lt;p&gt;Three weeks of vibe coding the new native Moolah and I&amp;rsquo;ve gone all-in. All my
data has been converted over to its iCloud backend. moolah-server is still
running but I haven&amp;rsquo;t been updating the data there so its getting more and more
out of date. At this point, I&amp;rsquo;m all in.&lt;/p&gt;
&lt;p&gt;As I write this Claude is happily deleting all the moolah-server backend
support. Apart from being a big milestone in terms of being committed to the new
native implementation, it also marks the first time that I can honestly say that
you can use Moolah and I won&amp;rsquo;t have any access to your data. Previously, data
was stored on the server. While I only ever stored a user ID from login, never
any name or other personal information, theoretically I could access the data in
the database. So while I expect I will remain the only actual user of Moolah,
it&amp;rsquo;s nice to know people &lt;em&gt;could&lt;/em&gt; use it without putting their privacy at risk.&lt;/p&gt;
&lt;p&gt;The bugs that were once so prevalent are now much more under control. In fact,
generally I&amp;rsquo;d say there aren&amp;rsquo;t many bugs, but there is a &lt;em&gt;lot&lt;/em&gt; of UI that lacks
polish or is just downright weird. UI tests are helping to get some of the UI
quirks under control, but UI design is definitely not AI&amp;rsquo;s strength.&lt;/p&gt;
&lt;p&gt;The more functional bugs on the other hand have been brought under control
through writing a series of guides and review agents - each focussed on a
specific area. There&amp;rsquo;s an iCloud sync reviewer, concurrency reviewer, UI
reviewer (good at ensuring everything has accessibility labels, not solving the
design problem) and a general code reviewer. Along with an insistence that it
fixes every issue raised by those reviewers, the codebase seems to be improving.&lt;/p&gt;
&lt;p&gt;Of course, I don&amp;rsquo;t actually write those guides - I just tell Claude to do deep
research on best practices and have it write the guide. Seems to work
surprisingly well&amp;hellip;&lt;/p&gt;
&lt;p&gt;The high level of test coverage also means that once I get it to fix a bug, it
generally stays fixed so quality is ratcheting up over time.&lt;/p&gt;
&lt;p&gt;The other technique that has been helpful for improving quality is to tell the
agent to step back and review the design to come up with a way to really
simplify it rather than just keep patching problems. The components that I&amp;rsquo;ve
gone through that process for have tended to stabilise pretty quickly.&lt;/p&gt;
&lt;p&gt;That said, progress has slowed down a lot. From the initial report on day 8 it
looked like multicurrency support was done. In practice, it&amp;rsquo;s only become
actually usable fairly recently with lots of details missing, particularly with
things just not being exposed through the UI. There&amp;rsquo;s a lot of rework happening
too - the data storage has moved from the most automatically managed storage and
sync system Apple provides, to doing sync manually and using SwiftData for
storage to now working through using SQLite manually (though the first external
dependency - GRDB). Nice improvements and nice that such large pieces can be
rewritten pretty quickly but it is showing the cost of not having understood
everything properly the first time around. Flip side, is we know a lot more
about requirements and performance trade offs now that I did at the start which
affects many of the choices anyway. It&amp;rsquo;s a pretty good sign that I feel
comfortable doing big migrations and architectural changes though. That&amp;rsquo;s partly
the relative ease with which it can be done - just through the AI at it rather
than spending time learning the new framework. Just as much though its a legacy
of having multiple backends and spending a lot of time ironing out bugs in the
initial iCloud backend - now there is a really great suite of tests that will
ensure this rewrite actually works.&lt;/p&gt;
&lt;p&gt;All an adventure, all a good learning experience and still a lot of fun.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>AI Agents Need Sandboxes, Not Permissions</title>
      <link>https://www.symphonious.net/2026/04/26/ai-agents-need-sandboxes-not-permissions/</link>
      <pubDate>Sun, 26 Apr 2026 18:19:00 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/04/26/ai-agents-need-sandboxes-not-permissions/</guid>
      <description>&lt;p&gt;AI agents like &lt;a href=&#34;https://claude.ai/claude-code&#34;&gt;Claude Code&lt;/a&gt; and &lt;a href=&#34;https://openai.com/codex/&#34;&gt;Codex&lt;/a&gt; try to be &amp;ldquo;safe&amp;rdquo; by limiting what the agent can do to a small set of operations and asking the user to approve everything else. In practice, the approval flow is so noisy and the requests so complex that it provides almost no real protection. We&amp;rsquo;ve reinvented Windows Vista&amp;rsquo;s never-ending UAC dialogs, and just like Vista, the only thing it actually trains users to do is click through.&lt;/p&gt;
&lt;p&gt;The permission model is fundamentally broken. The fix isn&amp;rsquo;t a better dialog — it&amp;rsquo;s a sandbox.&lt;/p&gt;
&lt;h2 id=&#34;the-approval-treadmill&#34;&gt;The Approval Treadmill&lt;/h2&gt;
&lt;p&gt;When the agent edits a file, the permission UI can show you the exact diff. Approving that is fine. The problem is that most of what an agent actually wants to do isn&amp;rsquo;t a clean file edit. It&amp;rsquo;s a multi-step bash invocation chaining &lt;code&gt;find&lt;/code&gt; into &lt;code&gt;xargs&lt;/code&gt; into &lt;code&gt;sed&lt;/code&gt;, or an inline Python script, or a &lt;code&gt;jq&lt;/code&gt; pipeline that the UI truncates after the first couple of hundred characters.&lt;/p&gt;
&lt;p&gt;Reviewing those properly takes real time. Reading a 30-line shell pipeline, working out what each step does, checking that nothing is being deleted that shouldn&amp;rsquo;t be, tracing where stdout is going to end up — that&amp;rsquo;s not a click, that&amp;rsquo;s a code review. And the agent fires off another one twenty seconds later.&lt;/p&gt;
&lt;p&gt;There are two outcomes. Either you do the careful review every time and spend more time vetting the agent than the agent saves you, or you stop reading and click yes. In practice everyone ends up at &amp;ldquo;click yes&amp;rdquo; within a couple of hours of real use (if that). That&amp;rsquo;s exactly where Vista left users — except now the thing on the other side of the dialog is actively generating novel commands you&amp;rsquo;ve never seen before.&lt;/p&gt;
&lt;h2 id=&#34;sandboxes-over-approvals&#34;&gt;Sandboxes Over Approvals&lt;/h2&gt;
&lt;p&gt;The alternative is to give the agent access to a safe but sufficient set of resources up front, and then let it work freely inside that boundary. No prompts, no review of every command — this is the box, do whatever you need inside it.&lt;/p&gt;
&lt;p&gt;&lt;a href=&#34;https://github.com/ajsutton/claude-docker&#34;&gt;claude-docker&lt;/a&gt; is one way to do this — run the agent in a container so the side effects are contained on a layer you can throw away. There are plenty of other options, many/most of them more polished and with much finer-grained controls over what the agent can reach. The built-in sandbox features in agent executors could work just as well too, but they all seem to default to &amp;ldquo;ask permission&amp;rdquo; rather than &amp;ldquo;deny&amp;rdquo;, and the sandbox they create is usually too small to actually get work done. So you end up turning the protections off or whitelisting things case by case, and you&amp;rsquo;re back where you started.&lt;/p&gt;
&lt;p&gt;The bar to set is: everything the agent has access to should be safe enough that you&amp;rsquo;d happily run it under &lt;code&gt;--dangerously-skip-permissions&lt;/code&gt; without a second thought. And the sandbox needs to be big enough that the agent rarely runs out of room to operate. When it does hit the wall, doing the missing step yourself outside the sandbox should be fine — because the alternative, loosening the sandbox so it doesn&amp;rsquo;t have to ask, defeats the whole point.&lt;/p&gt;
&lt;h2 id=&#34;hooks-arent-a-security-boundary&#34;&gt;Hooks Aren&amp;rsquo;t a Security Boundary&lt;/h2&gt;
&lt;p&gt;A tempting middle ground is to use hooks to inspect the bash commands the agent is about to run and reject the dangerous ones. This seems like a terribly insecure approach because there are too many ways to write the same operation. &lt;code&gt;git push&lt;/code&gt;, &lt;code&gt;git -c remote.origin.pushurl=… push&lt;/code&gt;, &lt;code&gt;git push origin HEAD:main&lt;/code&gt;, a shell variable expansion that hides the word entirely, a Python script that calls the GitHub API directly, a local helper invoked through &lt;code&gt;bash -c&lt;/code&gt;. A hook that searches for &lt;code&gt;git push&lt;/code&gt; will miss most of these, and a malicious agent — or a benign agent that has been prompt-injected — will find a route through.&lt;/p&gt;
&lt;p&gt;Hooks are still useful, just not as security. If you want the agent to be able to push but want a confirmation step on the way out, a hook will catch the common case and that&amp;rsquo;s a perfectly nice workflow as long as you don&amp;rsquo;t mind it failing every now and then. The mistake is treating that workflow nicety as a boundary that holds against an adversary. Anything that needs a real boundary needs to live at the sandbox layer, not in a regex over command lines.&lt;/p&gt;
&lt;h2 id=&#34;fine-grained-tokens&#34;&gt;Fine-Grained Tokens&lt;/h2&gt;
&lt;p&gt;For sandboxes to be genuinely useful, the things the agent reaches into them — GitHub, CI, package registries, chat — need to support narrow credentials. &amp;ldquo;Read&amp;rdquo; and &amp;ldquo;write&amp;rdquo; aren&amp;rsquo;t enough resolution. What can be read, what can be written?&lt;/p&gt;
&lt;p&gt;I want to be able to give an agent a token that can create issues but not push to protected branches. Or one that can read CircleCI build results from a couple of specific projects but not change their settings or look at unrelated repos. Or a Slack token that can post to one channel and nowhere else. The granularity I&amp;rsquo;d give a junior engineer for a focused task is roughly the granularity I want to give an agent — and that&amp;rsquo;s much finer than most platforms expose today.&lt;/p&gt;
&lt;p&gt;GitHub&amp;rsquo;s fine-grained personal access tokens are a step in the right direction, but the granularity is uneven and the UI is painful enough that most people fall back to broad-scope tokens anyway. If we want agents to operate inside a tight sandbox without the human constantly stepping in to do the bits the sandbox can&amp;rsquo;t reach, the platforms the agent talks to need to make narrow, scoped credentials easy to issue and obvious to use.&lt;/p&gt;
&lt;p&gt;The &amp;ldquo;ask permission for everything&amp;rdquo; model isn&amp;rsquo;t going to get fixed with a nicer dialog — it&amp;rsquo;s the wrong shape. Real protection looks like a sandbox the agent can move freely inside and credentials that can&amp;rsquo;t do much damage even if the agent goes off the rails. The sandbox tooling is rough today and the credential story is worse, but those are the levers worth pushing on, not another permission prompt.&lt;/p&gt;</description>
    </item>
    
    <item>
      <title>Moolah Diaries: Vibe Coding a New Moolah</title>
      <link>https://www.symphonious.net/2026/04/13/moolah-diaries-vibe-coding-a-new-moolah/</link>
      <pubDate>Mon, 13 Apr 2026 19:00:00 +1000</pubDate>
      <guid>https://www.symphonious.net/2026/04/13/moolah-diaries-vibe-coding-a-new-moolah/</guid>
      <description>&lt;p&gt;I got Claude to &lt;a href=&#34;https://www.symphonious.net/2026/04/13/moolah-vibe-coding/&#34;&gt;crunch the numbers&lt;/a&gt; on the moolah-native project — eight days of vibe coding a SwiftUI replacement for my long-running personal finance tracker, with zero Swift experience. The stats are interesting but they don&amp;rsquo;t really capture what the experience has been like, so here&amp;rsquo;s the human version.&lt;/p&gt;
&lt;h3 id=&#34;skipping-the-learning-curve&#34;&gt;Skipping the Learning Curve&lt;/h3&gt;
&lt;p&gt;Using AI to just skip the initial learning curve of a new language and platform has been really freeing. I don&amp;rsquo;t know Swift. I don&amp;rsquo;t know SwiftUI. Normally I&amp;rsquo;d spend a ton of time working out the basic setup and finding what patterns/libraries to use before I could do anything useful and that&amp;rsquo;s often enough to kill the motivation entirely. Instead I was working on actual domain problems from day one — straight to the interesting part. The learning curve didn&amp;rsquo;t go away, I just got to avoid it. Will this come back to bite me? Quite possibly.&lt;/p&gt;
&lt;h3 id=&#34;brainstorming-with-a-pretty-smart-duck&#34;&gt;Brainstorming with a Pretty Smart Duck&lt;/h3&gt;
&lt;p&gt;The brainstorming has been the standout. Claude Code has a &lt;a href=&#34;https://github.com/nicobailon/superpowers-skills&#34;&gt;brainstorming skill&lt;/a&gt; that walks you through refining designs interactively, and it&amp;rsquo;s been great. It found data sources for exchange rates, stock prices, and crypto prices that I&amp;rsquo;d missed even after searching pretty thoroughly myself. That opened up features I&amp;rsquo;d been uncertain about.&lt;/p&gt;
&lt;p&gt;But the bigger win was at the design level. I&amp;rsquo;d previously looked at adding multi-currency support and even wrote a lot of the code but never really felt the design hit the complexity/benefit trade off. Working through the brainstorming process came up with a multi-leg transaction model that handles multicurrency much more cleanly &lt;em&gt;and&lt;/em&gt; simplifies complexity around the existing transfer support. It&amp;rsquo;s rubber ducking, but with a duck reasonably smart duck (that has opinions and can pull up documentation you didn&amp;rsquo;t know existed).&lt;/p&gt;
&lt;h3 id=&#34;the-cost-problem&#34;&gt;The Cost Problem&lt;/h3&gt;
&lt;p&gt;AI usage limits are incredibly annoying. A Claude Pro plan is basically useless for anything beyond toy code — you&amp;rsquo;ll hit limits within an hour of real work. Gemini&amp;rsquo;s free tier is surprisingly capable for casual use, but if you want to actually sustain any kind of productivity you wind up needing Claude&amp;rsquo;s Max 20x plan and that&amp;rsquo;s a lot of money for a side project which may be dormant for long periods. For a business the cost is a no-brainer - it can make expensive engineers way more productive so pays for itself. For pesonal use when you have a family and look at costs in AUD, not so much.&lt;/p&gt;
&lt;h3 id=&#34;the-bugs&#34;&gt;The Bugs&lt;/h3&gt;
&lt;p&gt;Oh my god, the bugs. The &lt;a href=&#34;https://www.symphonious.net/2026/04/13/moolah-vibe-coding/&#34;&gt;analysis post&lt;/a&gt; puts the fix rate at 31% compared to 2.6% for the hand-written server. That tracks and maybe understates it. I&amp;rsquo;m pretty damn impressed with the low bug rate for moolah-server though given its a side project and I always felt like I was cutting corners and not writing tests that I should have.&lt;/p&gt;
&lt;p&gt;Today was the worst day yet. At some point recently, AI had split iCloud sync profiles into separate CloudKit zones but they were actually still in a single zone, overlapping with each other. We&amp;rsquo;d talked about that and it was super confident it would work even when pressed on it and told to do deep research. Fixing that required migrating to a lower-level sync API (CKSyncEngine), which then triggered a cascade of stability problems and has taken pretty much the whole day to fix.&lt;/p&gt;
&lt;p&gt;AI has the same instinct as a lot of engineers (including plenty of senior ones): when something breaks, keep throwing bandaids at the symptoms. Track more state. Add another deduplication pass. Cache another field. Each fix is locally reasonable but the complexity just ratchets up. I wound up having to step in and push it to throw away all the state tracking and just model the problem better so it automatically did the right thing. Less code, fewer bugs, simpler to reason about. But AI won&amp;rsquo;t get there on its own — it optimises for making the current test pass, not for finding a design where the test wouldn&amp;rsquo;t have failed in the first place.&lt;/p&gt;
&lt;h3 id=&#34;shipping-is-still-hard&#34;&gt;Shipping is Still Hard&lt;/h3&gt;
&lt;p&gt;I&amp;rsquo;ve started entering transactions through the SwiftUI app but I&amp;rsquo;m still running the original moolah-server as the backend because I don&amp;rsquo;t trust the iCloud sync layer. Today is a pretty good example of why.&lt;/p&gt;
&lt;p&gt;There&amp;rsquo;s also a subtler problem: vibe coding makes it very tempting to just keep churning out features because it&amp;rsquo;s fun to watch AI produce things. But every new feature is more unvalidated code, and that works against ever actually shipping. At some point you have to stop adding and start using. But I have nearly 15 years of historic data at risk and detecting subtle corruption or a few messed up transactions can be really hard.&lt;/p&gt;
&lt;h3 id=&#34;its-fun-though&#34;&gt;It&amp;rsquo;s Fun Though&lt;/h3&gt;
&lt;p&gt;Coding was my hobby before it became my job. I still enjoy the job but it&amp;rsquo;s serious work with serious consequences. Vibe coding has brought back some of the fun of just playing with technology for its own sake — partly the quick results, but also just experimenting with a powerful new tool and learning how to get the most out of it.&lt;/p&gt;
&lt;p&gt;The code quality concerns are real, the costs are high, and the bugs are maddening. But I built a functional multi-platform finance app in eight days with no platform experience, and I had a good time doing it.&lt;/p&gt;</description>
    </item>
    
  </channel>
</rss>