[{"data":1,"prerenderedAt":241},["ShallowReactive",2],{"notes":3,"profile":52,"note-developer-tools-for-normal-teams":239},[4,13,19,26,34,40,46],{"slug":5,"title":6,"topic":7,"status":8,"published":9,"summary":10,"minutes":11,"href":12},"a-notes-site-that-looks-like-my-terminal","A notes site that looks like my terminal","Experiments","published","2026-11-17","A notes site drawn as a terminal draws Markdown, because that is the screen I look at all day, and because agents read the same file.",2,"\u002Fnotes\u002Fa-notes-site-that-looks-like-my-terminal",{"slug":14,"title":15,"topic":7,"status":8,"published":9,"summary":16,"minutes":17,"href":18},"working-with-agents-by-ear","Working with agents by ear","After eye surgery I couldn't look at a screen for long, so I steered my coding agents by voice. What made that usable, and the plugin that came out of it.",3,"\u002Fnotes\u002Fworking-with-agents-by-ear",{"slug":20,"title":21,"topic":22,"status":8,"published":23,"summary":24,"minutes":17,"href":25},"one-person-many-agents","One person, many agents, and a phone","Systems","2026-10-01","How I run many coding agents across a Mac and a rented Linux box from my phone, and the parts you need to build the same.","\u002Fnotes\u002Fone-person-many-agents",{"slug":27,"title":28,"topic":29,"status":8,"published":30,"updated":9,"summary":31,"minutes":32,"href":33},"developer-tools-for-normal-teams","Developer tools for normal teams","AI","2026-09-24","Every team using AI is about to hit problems software teams solved years ago. The fix is the same four tools under other names.",4,"\u002Fnotes\u002Fdeveloper-tools-for-normal-teams",{"slug":35,"title":36,"topic":7,"status":8,"published":37,"updated":9,"summary":38,"minutes":11,"href":39},"jev-tiny-models-strange-interfaces","Jev, tiny models and strange new interfaces","2026-09-23","Small, fast models that pick one option from a fixed set don’t chat. They decide, and that invites interfaces that don’t look like chat at all.","\u002Fnotes\u002Fjev-tiny-models-strange-interfaces",{"slug":41,"title":42,"topic":43,"status":8,"published":37,"updated":9,"summary":44,"minutes":11,"href":45},"when-software-becomes-an-actor","What finance looks like when software becomes an actor","Finance","When software can act on money, not just record it, the hard questions stop being technical and become about permission and accountability.","\u002Fnotes\u002Fwhen-software-becomes-an-actor",{"slug":47,"title":48,"topic":49,"status":8,"published":37,"updated":9,"summary":50,"minutes":11,"href":51},"ai-makes-software-cheaper","AI makes software cheaper. Systems thinking becomes more valuable.","Software","Models make code cheap to write. They don’t make systems cheap to own, and that gap is where the valuable work moves.","\u002Fnotes\u002Fai-makes-software-cheaper",{"name":53,"meta":54,"statement":55,"lede":58,"description":59,"origin":60,"role":61,"work":62,"trajectory":102,"notes":131,"about":134,"contact":170,"home":184},"Chris Wijnia","Engineer · Co-founder",{"text":56,"accent":57},"I build software that holds up.","holds up.","I work out what’s worth building next.","Dutch engineer and co-founder. I work out what’s worth building next.","Netherlands","Product \u002F R&D",{"title":63,"accent":64,"outro":65,"projects":66},"A few things I’ve worked on.","worked on.","The work has changed a lot over the years. The habit of making things hasn’t.",[67,73,79,85,91,96],{"no":68,"name":69,"area":70,"period":71,"slug":72},"01","AI systems","Agents \u002F Automation \u002F Developer tools","2022–Now","ai-systems",{"no":74,"name":75,"area":76,"period":77,"slug":78},"02","1GRAM","Wallet software","2020–2021","1gram",{"no":80,"name":81,"area":82,"period":83,"slug":84},"03","MVRDV","Digital platform \u002F Architecture","2015–2019","mvrdv",{"no":86,"name":87,"area":88,"period":89,"slug":90},"04","JIJ & Overijssel","Public participation platform","2014–2018","jij-overijssel",{"no":92,"name":93,"area":94,"period":95,"slug":93},"05","100-100-100","Sustainability campaign platform","2014–2019",{"no":97,"name":98,"area":99,"period":100,"slug":101},"06","Archive","Design \u002F Web \u002F Software","2007–2022","archive",{"title":103,"eras":104},"Moving down the stack as the web evolved.",[105,109,113,116,120,124,127],{"year":106,"name":107,"line":108},"2005","Design","Learning the web by making things.",{"year":110,"name":111,"line":112},"2013","Product","Design became code.",{"name":114,"line":115},"Engineering","Code became systems.",{"year":117,"name":118,"line":119},"2018","Wallets","Software for a new kind of money.",{"year":121,"name":122,"line":123},"2021","Financial infra","Building systems where software meets regulation.",{"year":125,"name":29,"line":126},"2022","From using models to orchestrating them.",{"year":128,"name":129,"line":130},"Now","Building","All of it, at once.",{"title":132,"intro":133},"Notes","Things I’m working on, learning or trying to understand.",{"title":135,"paragraphs":136,"byte":140,"company":147},"I’ve been making things on computers for most of my life.",[137,138,139],"I grew up in the Netherlands and built my first websites at thirteen.","I like small teams, ambitious projects, good engineering and people who can disagree without making it tedious. I still enjoy opening an empty repository and figuring something out.","I work internationally. Outside work I train Muay Thai and like [chess](\u002Flab\u002Fchess), which I played competitively in Zwolle and taught to primary-school children there. I also build things that occasionally have no sensible commercial purpose.",{"headline":141,"body":145,"prompt":146},[142,143,144],"Eight bits","make a","letter.","I started in design, placing pixels one at a time until a shape read as a word. Software is built the same way, from the smallest unit up.","Flip a bit. Make a letter.",{"headline":148,"body":153,"principles":154},[149,150,151,152],"A company","is a system","with people","inside.","A company runs on what good software runs on: clear states, clean interfaces, and nothing that only works as long as someone remembers.",[155,158,161,164,167],{"name":156,"line":157},"Resolution is a choice","Decide how much detail a problem deserves, then commit to it.",{"name":159,"line":160},"Two states beat five","If a feature cannot say yes or no, it is not finished.",{"name":162,"line":163},"The ledger balances","Every action leaves a record, and the records add up.",{"name":165,"line":166},"The model proposes","A person signs. The system always knows which is which.",{"name":168,"line":169},"Every pixel on purpose","Including the ones nobody will ever notice.",{"title":171,"accent":172,"intro":173,"links":174},"Making something interesting?","interesting?","For collaborations, technical discussions, or a good reason I haven’t thought of.",[175,178,181],{"label":176,"href":177},"LinkedIn","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fchris-wijnia-1364a29b",{"label":179,"href":180},"GitHub","https:\u002F\u002Fgithub.com\u002Fcwdx",{"label":182,"href":183},"X","https:\u002F\u002Fx.com\u002FChristianWijnia",{"hero":185,"chapters":192,"work":229,"contact":233},{"headline":186,"lede":190,"cue":191},[187,188,189],"Every system","is one bit,","repeated.","I’m a Dutch engineer and co-founder, working on financial software and AI systems.","Scroll to scale up",[193,204,217],{"id":194,"power":195,"name":196,"label":197,"formation":194,"cell":198,"headline":199,"body":202,"prompt":203},"bit","2^0","Bit","The bit · 1 state of 2",16,[200,201],"Yes.","Or no.","Approve the payment or hold it. Trust the answer or ask again. Ship the pixel or move it one to the left. Everything I build ends in a decision small enough to fit in one bit. The work is making sure it is the right one.","Flip the bit. This page has two states.",{"id":205,"power":206,"name":207,"label":208,"formation":205,"cell":32,"ink":209,"headline":210,"body":215,"prompt":216,"case":78},"ledger","2^10","Ledger","The ledger · financial software",true,[211,212,213,214],"Money is","the least","forgiving","data type.","In financial software, probably right is wrong. No rounding error is too small to matter, and no state is allowed to drift. Every debit meets its credit. Every action leaves a record that adds up, today and in the audit years from now.","Simulated ledger",{"id":218,"power":219,"name":220,"label":221,"formation":218,"cell":11,"headline":222,"body":227,"prompt":228,"note":47},"weights","2^20","Weights","The weights · AI systems",[223,224,225,226],"A million","decisions,","none made","on purpose.","A model is a vast field of numbers nobody set by hand. Useful, and never quite certain. I build the part around it: where it may act alone, where it has to ask, and how anyone can check afterwards what it did and why.","Confidence",{"label":230,"headline":231},"Work · 2007–now",[232],"Work.",{"label":234,"headline":235,"intro":238},"Back to one bit",[236,237],"Send","one bit.","Building something where a wrong answer costs money, or where a machine’s answer has to be trusted? Tell me in a line. I will answer in more than one bit.",{"slug":27,"title":28,"topic":29,"status":8,"published":30,"updated":9,"summary":31,"minutes":32,"href":33,"html":240},"\u003Cp>Every team that uses AI seriously runs into the same wall, and it isn’t the price of tokens. Once several agents are doing real work, nobody can say for certain what they did or whether it was right.\u003C\u002Fp>\n\u003Cp>Software teams have had that problem for decades, with people instead of agents, and they solved it with a handful of tools. I think those tools are about to leave engineering. This is the argument I made at an AI meetup. The slides are at the end.\u003C\u002Fp>\n\u003Ch2 id=\"the-wall-is-trust\" data-no=\"01 \u002F 09\" data-hud=\"01 The wall is trust\">The wall is trust\u003C\u002Fh2>\n\u003Cp>Starting an agent is easy. Anyone can point one at an inbox or a spreadsheet in an afternoon.\u003C\u002Fp>\n\u003Cp>The hard part comes afterwards. An agent rarely fails loudly. It summarises a little less carefully, skips a field in the CRM, drafts a proposal that is nearly right. If nothing checks the work and nothing records it, you find out weeks later, from a customer.\u003C\u002Fp>\n\u003Cp>So the first question I’d ask about any agent is “how would I know?”\u003C\u002Fp>\n\u003Ch2 id=\"how-a-small-team-runs-millions-of-lines\" data-no=\"02 \u002F 09\" data-hud=\"02 How a small team runs millions of lines\">How a small team runs millions of lines\u003C\u002Fh2>\n\u003Cp>The codebase I work in every day runs to millions of lines across dozens of apps and services, and a small team maintains it. No one person holds all of it in their head.\u003C\u002Fp>\n\u003Cp>It works because the agents that do much of the work operate inside a few rules we wrote down, and because nothing they produce lands without being checked. The rules are written playbooks, which we call skills. The work is split between specialised agents for planning, coding, reviewing and shipping. None of that is exotic. It’s mostly the standard toolkit of any well-run engineering team, applied to agents instead of people.\u003C\u002Fp>\n\u003Ch2 id=\"four-tools-under-other-names\" data-no=\"03 \u002F 09\" data-hud=\"03 Four tools under other names\">Four tools under other names\u003C\u002Fh2>\n\u003Cp>Each of these exists in software because many hands working in one system break things in ways nobody notices until later. Agents are many hands.\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Checks the agent can run itself.\u003C\u002Fstrong> Developers have tests: code that fails when something breaks. For everyone else it’s a checklist, a sample to compare against, or a number that has to move. An agent that can check its own work improves without you. One that can’t makes you the test.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>A record of everything.\u003C\u002Fstrong> Developers have version history: every change, who made it, and a way back. For everyone else it’s Linear, Drive or Notion, as long as the agent writes where people already read.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Know-how written down once.\u003C\u002Fstrong> Developers have runbooks. For agents, the same idea is a skill: a short instruction loaded when the task needs it. The useful ones come from watching an agent fail at the same thing twice.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Gates sized to the mistake.\u003C\u002Fstrong> Developers have code review. For everyone else it’s approval: light for a draft email, heavier for a payment or a contract.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cfigure class=\"mermaid-embed not-prose\">\u003Cfigcaption class=\"mermaid-state mono-label\">Diagram\u003C\u002Ffigcaption>\u003Cpre class=\"mermaid-src\">flowchart LR\n  D1[&quot;Tests&quot;] --&gt; E1[&quot;A checklist, a sample&lt;br\u002F&gt;or a number that moves&quot;]\n  D2[&quot;Version history&quot;] --&gt; E2[&quot;Linear, Drive or Notion&quot;]\n  D3[&quot;Runbooks&quot;] --&gt; E3[&quot;A skill&quot;]\n  D4[&quot;Code review&quot;] --&gt; E4[&quot;Approval sized&lt;br\u002F&gt;to the mistake&quot;]:::lit\n\u003C\u002Fpre>\u003C\u002Ffigure>\n\u003Cp>A team can pick up all four habits without hiring a developer.\u003C\u002Fp>\n\u003Ch2 id=\"keeping-track-is-the-one-to-start-with\" data-no=\"04 \u002F 09\" data-hud=\"04 Keeping track is the one to start with\">Keeping track is the one to start with\u003C\u002Fh2>\n\u003Cp>If I had to pick one, it’s the record. Every agent, whatever it does, writes to the same place: what it did, and whether its check passed.\u003C\u002Fp>\n\u003Cp>That’s also how you catch the quiet failure. A shared log turns “the agent seems fine” into something you can see. Work that isn’t tracked can’t be checked later, so nobody can tell whether it went wrong.\u003C\u002Fp>\n\u003Ch2 id=\"each-step-needs-a-new-guardrail\" data-no=\"05 \u002F 09\" data-hud=\"05 Each step needs a new guardrail\">Each step needs a new guardrail\u003C\u002Fh2>\n\u003Cp>Boris Cherny, who built Claude Code, \u003Ca href=\"https:\u002F\u002Fx.com\u002Fbcherny\u002Fstatus\u002F2077929379661844559\" target=\"_blank\" rel=\"noopener\">describes five steps of AI adoption\u003C\u002Fa>: gated, then assisted (about one agent per person), parallel (about ten), supervised autonomy (about a hundred), and AI-native (a thousand or more). His point, which matches what I’ve seen, is that tokens alone don’t move a team up a step. Removing the next bottleneck and building the next guardrail does.\u003C\u002Fp>\n\u003Cp>The four tools above are those guardrails. Most teams outside engineering are on the first or second step, and the jump to the third is where they start to matter. At ten agents per person, nobody can supervise every output by eye.\u003C\u002Fp>\n\u003Ch2 id=\"one-install-for-the-whole-team\" data-no=\"06 \u002F 09\" data-hud=\"06 One install for the whole team\">One install for the whole team\u003C\u002Fh2>\n\u003Cp>Habits spread badly by memo. What worked for us was packaging them. The team gets one plugin for Claude Cowork, installed once, with no terminal involved. It holds the skills, the connections to our own systems and the checks:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Skills.\u003C\u002Fstrong> For example, an inbox assistant that learns your tone from your sent mail and drafts replies on a schedule.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Connectors.\u003C\u002Fstrong> Links to the company’s docs, CRM and data, so a question in plain words gets an answer from the real source.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Checks.\u003C\u002Fstrong> For example, a fact-checker that re-derives every claim in a draft before it goes out.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>The plugin is built from what we actually run, so when someone adds a skill or a connection, everyone has it on the next update. Nobody maintains a separate list, so there is no list to go stale.\u003C\u002Fp>\n\u003Ch2 id=\"can-normal-teams-run-this\" data-no=\"07 \u002F 09\" data-hud=\"07 Can normal teams run this?\">Can normal teams run this?\u003C\u002Fh2>\n\u003Cp>The meetup asked. Yes. None of the four tools requires writing code, and a plugin installs them for everyone at once. Agents break things quietly when there’s no check and no record. With both, a problem shows up in the log the same day.\u003C\u002Fp>\n\u003Ch2 id=\"the-slides\" data-no=\"08 \u002F 09\" data-hud=\"08 The slides\">The slides\u003C\u002Fh2>\n\u003Cdiv class=\"deck-embed not-prose\" data-deck=\"developer-tools-for-normal-teams\">\u003Cp class=\"mono-label\">\u003Ca href=\"\u002Fdecks\u002Fdeveloper-tools-for-normal-teams\">Open the deck ►\u003C\u002Fa>\u003C\u002Fp>\u003C\u002Fdiv>\n\u003Ch2 id=\"open-questions\" data-no=\"09 \u002F 09\" data-hud=\"09 Open questions\">Open questions\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>How small can the record be before it stops being useful? A spreadsheet works for one team; I don’t know where it stops working.\u003C\u002Fli>\n\u003Cli>Who writes the skills in a team with no engineers, and who decides when one is wrong?\u003C\u002Fli>\n\u003Cli>Checks are easy for numbers and hard for judgement. What’s the checklist for “is this a good proposal”?\u003C\u002Fli>\n\u003C\u002Ful>\n",1795075804217]