# Better Than Good

> We solve problems with strategy, design, code and AI, with two generations of experience helping organizations become Better Than Good.

## Pages

- [About Better Than Good](https://betterthangood.xyz/): Introduction to Better Than Good, Garrett Winder, and Iain Harper
- [Blog](https://betterthangood.xyz/blog/): All blog posts on strategy, technology and AI
- [RSS feed](https://betterthangood.xyz/feed.xml): Feed of every blog post

## Blog posts

- [Foomers vs doomers: skilled incompetence and the AI tug of war](https://betterthangood.xyz/blog/foomers-vs-doomers/): AI-assisted coding has split organisations into boosters and sceptics. The schism isn't cultural — it's a feedback loop broken in four diagnosable places, and each one can be repaired.
- [Forward Deployed Engineers: AI’s hired guns](https://betterthangood.xyz/blog/hired-guns/): Every major AI lab is spending billions to embed forward-deployed engineers inside customers. What the FDE boom really admits about AI’s maturity — and the vendor lock-in it quietly rebuilds.
- [The sorting machine: recruitment’s race to the bottom](https://betterthangood.xyz/blog/the-sorting-machine/): AI now scores the resumes AI wrote, and qualified people vanish on a wrong synonym or a six-month gap. How recruitment’s sorting machine breaks — and the fix.
- [Weighing smoke: why GEO dashboards are mostly useless](https://betterthangood.xyz/blog/weighing-smoke/): GEO dashboards launder daily noise into a decimal-pointed visibility score over an invented basket of prompts. What actually tracks, and the cheap alternative.
- [Another nice mess](https://betterthangood.xyz/blog/another-nice-mess/): Twenty years of sensibly chosen SaaS tools have left small businesses with a patchwork data estate. AI makes fixing it urgent—and a thin orchestration layer is how you do it without the enterprise price tag.
- [The state and the machine](https://betterthangood.xyz/blog/the-state-and-the-machine/): The US government forced Anthropic to switch off Fable 5 and Mythos 5 over a weekend — a sign that self-improving AI has become a matter of state power.
- [We have ways of making you pay](https://betterthangood.xyz/blog/we-have-ways-of-making-you-pay/): The frontier labs are switching the meter back on, and rising bills are only half the story. The other half is that nobody, buyer or seller, can say what a unit of AI work costs.
- [Bloated: how chat made you fat](https://betterthangood.xyz/blog/bloated-how-chat-made-you-fat/): The time chat tools save you gets charged to your readers and padded out across pages they didn't need. Why models pad, and the only fix that actually trims a document.
- [Apple’s bicycle without a chain](https://betterthangood.xyz/blog/apples-bicycle-without-a-chain/): Private Cloud Compute is an engineering marvel; Image Playground produces hooved tacos. The chain Apple has been forging beneath the bicycle, and what WWDC 2026 must show.
- [Weeknotes vol. 17: business, schmizness](https://betterthangood.xyz/blog/weeknotes-17/): Iain joins full time, we’re solving operational problems with Claude and I’ve been turning folders, files and rules into automated work.
- [The ten trillion dollar gamble](https://betterthangood.xyz/blog/ten-trillion-dollar-gamble/): OpenAI's CFO floated a federal 'backstop' for $1.4 trillion of compute contracts. The bets, mechanics, and unknowns behind the AI buildout.
- [Never talk about goblins](https://betterthangood.xyz/blog/never-talk-about-goblins/): Every line in a production AI system prompt is a scar from a past failure. OpenAI's 'never talk about goblins' is the funny one — and a warning.
- [Read the frickin’ manual: the end of the user interface](https://betterthangood.xyz/blog/read-the-frickin-manual/): The user interface was doing work no one wrote down. When agents replace users, all that implicit knowledge has to be documented — or it disappears.
- [Attention is all you ever needed](https://betterthangood.xyz/blog/attention-is-all-you-ever-needed/): Pareto's 80/20 rule was never a law of commerce — it described mid-century factories and broadcast TV. AI is now dismantling both.
- [The cost of everything and value of nothing](https://betterthangood.xyz/blog/cost-everything-value-nothing/): AI pricing in 2026 looks like mobile data in 2008 — messy meters, clever tiers, and a predictable drift toward flat-rate predictability.
- [Go to the actual place and see the actual thing](https://betterthangood.xyz/blog/go-see-actual-work-ai-lean-ohno-circle/): Most AI projects fail because companies skip a simple rule: go and see the work. Learn how Toyota’s lean principles—like the Ohno Circle, value stream mapping, and kaizen—offer a proven framework for successful AI implementation in service organisations.
- [Climbing the Claude ladder: from prompting to orchestrating](https://betterthangood.xyz/blog/prompting-to-orchestrating/): Most people use Claude at the prompting level and plateau there. Here's the four-rung ladder from single prompts to agent orchestration, and why each level compounds on the last.
- [The path to an agent-first web](https://betterthangood.xyz/blog/path-agent-first-web/): AI agents are reshaping how commerce works online. Here's what the protocol wars, Amazon's walled garden, and WebMCP mean for every site owner.
- [Snake oil, SEO, and the GEO chimera](https://betterthangood.xyz/blog/geo-practice-versus-snake-oil/): GEO has no SERP, no positions, and 40-60% variance with no content change. What the evidence supports, and what the salesmen invented.
- [Automating your marketing 01: Paid Search Ads](https://betterthangood.xyz/blog/automating-paid-search-ads/): A practical guide to automating Google Search Ads with AI — what to automate, what to supervise, and what to keep in human hands.
- [Why AI models hallucinate](https://betterthangood.xyz/blog/why-ai-models-hallucinate/): OpenAI's own research shows evaluation systems actively incentivise hallucination. Why rewarding guesses over honest uncertainty makes the problem structural.
- [Received wisdom: classic frameworks under AI pressure 01: David C Baker](https://betterthangood.xyz/blog/ai-pressure-david-baker/): David C Baker's positioning frameworks shaped agency strategy for thirty years. AI is compressing the knowledge barrier they relied on.
- [The trust problem that you already solved](https://betterthangood.xyz/blog/trust-problem-already-solved/): You already trust opaque systems that produce predictable output. AI-generated code is the same problem — and the same framework applies.
- [Received wisdom: classic frameworks under AI pressure 02: Crabtree’s LER](https://betterthangood.xyz/blog/ai-pressure-crabtree-ler/): AI tools that boost output without appearing in labour costs are distorting Crabtree's LER. The fix: widen what counts as production cost.
- [The production agent stack for sensitive environments: a field guide for 2026](https://betterthangood.xyz/blog/production-agent-stack-2026/): An opinionated guide to the 2026 production agent stack — orchestration, security, observability, and cost control for sensitive environments.
- [Yes, the models got dumber](https://betterthangood.xyz/blog/models-getting-dumber/): GPT-4 lost 95 points on prime number detection in three months. Model degradation via quantisation, routing changes, and silent updates is real and measurable.
- [The sunk cost of being good at something](https://betterthangood.xyz/blog/sunk-cost-expertise/): AI whataboutism uses real concerns as reasons for inaction. The psychology of status quo bias, sunk cost, and why expertise can become its own trap.
- [The flatness of the machine](https://betterthangood.xyz/blog/ai-writing-has-no-voice/): AI prose is fluent, frictionless, and hollow. Why next-token prediction, RLHF, and model collapse are systematically and increasingly optimized against voice.
- [Meta and Stripe want you to buy things from ads again](https://betterthangood.xyz/blog/meta-stripe-ad-checkout/): Meta and Stripe built native checkout inside Facebook ads — less purchase friction, clearer attribution, and a challenge to Amazon's one-click.
- [Meta’s GEM: what the largest ads foundation model means for your marketing](https://betterthangood.xyz/blog/meta-gem-ads-foundation-model/): Meta’s GEM is the largest foundation model ever built for ad recommendation. What its sequence learning architecture means for marketers and ad targeting.
- [The narrow window for probabilistic agents](https://betterthangood.xyz/blog/probabilistic-agents-narrow-window/): Most businesses buying AI agents need better plumbing, not probabilistic reasoning. Know where genuine uncertainty demands these tools.
- [Your org chart is not your AI strategy](https://betterthangood.xyz/blog/org-chart-not-ai-strategy/): Shadow AI is an organisational design problem, not a security one. Why IT shouldn't own AI strategy and what effective adoption actually requires.
- [Software was never meant to last forever](https://betterthangood.xyz/blog/software-never-meant-to-last-forever/): The economics of enterprise SaaS are shifting. AI is making software disposable—migration is cheaper, building is faster, and the future belongs to composable systems.
- [The machine that improves the machine](https://betterthangood.xyz/blog/alphaevolve-recursive-improvement/): Google's AlphaEvolve uses AI to discover better algorithms — and some improvements are already accelerating the models that power it.
- [The vibe coding spectrum: from weekend hacks to the dark factory](https://betterthangood.xyz/blog/vibe-coding-spectrum-weekend-hacks-dark-factory/): From Karpathy's weekend hacks to StrongDM's dark factory where no human writes or reviews code. How vibe coding evolved into software factories in one year.
- [A $10K Mac Studio won’t replace your API bill](https://betterthangood.xyz/blog/mac-studio-wont-replace-api/): A $10K Mac Studio running open-source models sounds liberating. But local inference still can't match frontier APIs for agentic, high-reasoning work. Here's why.
- [Claude Opus 4.6 just shipped agent teams. But can you trust them?](https://betterthangood.xyz/blog/claude-opus-46-agent-teams-trust/): Claude Opus 4.6 introduces agent teams for parallel coding tasks. But multi-agent coordination creates new security vulnerabilities the industry hasn't solved.
- [Out of context: strategies for managing agent memory](https://betterthangood.xyz/blog/agent-memory-strategies/): Context windows are marketed as vast, but models reliably use only half their capacity. Here are practical strategies for those limits.
- [Escaping prototype purgatory: where is AWS for AI agents?](https://betterthangood.xyz/blog/escaping-prototype-purgatory/): Building production AI agents is a mess of security, identity, memory, and observability. Could a managed platform solve it the way AWS solved cloud infrastructure?
- [The Hot Mess: large AI models and the scaling mirage](https://betterthangood.xyz/blog/scaling-mirage-model-errors/): As AI models scale up, their remaining errors become more random and unpredictable. The implications for autonomous agents are significant.
- [Tooling around: letting agents do stuff is hard](https://betterthangood.xyz/blog/agent-tooling-mcp-security/): MCP became the default way to connect AI models to tools faster than security could keep up. What's actually involved in deploying MCP servers safely.
- [Building a simple agent with Claude](https://betterthangood.xyz/blog/building-simple-claude-agent/): A practical walkthrough of building a sales prep agent using Claude's tool use API — three tools, one system prompt, and immediate ROI.
- [How Claude Code and Cowork talk to your other systems](https://betterthangood.xyz/blog/claude-code-mcp-connections/): Claude Code and Cowork use the Model Context Protocol to connect AI to external tools. Here is how the plumbing works and why security matters.
- [Security for production AI agents in 2026](https://betterthangood.xyz/blog/security-production-ai-agents/): A practical guide to securing AI agents in production. Defence-in-depth strategies combining deterministic validators, LLM-based evaluation, human oversight, and observability.
- [In the jungle: a reality check on AI agents](https://betterthangood.xyz/blog/reality-check-ai-agents/): AI agents promise what RPA promised a decade ago. Here's why most agentic AI projects are failing and what the automation playbook teaches us about building ones that work.
- [AI governance: between the committee and the catastrophe](https://betterthangood.xyz/blog/ai-governance-proportional/): Most organisations govern AI badly, either strangling innovation with bureaucracy or shipping models with no oversight at all.
- [AI slop: psychology, history, and the problem of the ersatz](https://betterthangood.xyz/blog/ai-slop-psychology-history-ersatz/): Why do people find AI-generated content so viscerally off-putting? The psychology of authenticity, historical parallels to the Luddites, and what it means for creative work.
- [Why AI agents keep forgetting things, and the race to fix it](https://betterthangood.xyz/blog/ai-agent-memory-race/): AI agents lose everything when context closes. New research tackles the memory problem with reinforcement learning and indexed archiving.
- [The missiles are the destination](https://betterthangood.xyz/blog/the-missiles-are-the-destination/): The missiles are the destination. Why problem-solving isn’t a phase you get through, it’s the actual work of building something valuable.
- [Fall back](https://betterthangood.xyz/blog/fall-back/): If everyone is rushing to be 'AI-powered,' are we all just leaping into a commodity trap? What digital agencies actually need to do to stay relevant without losing what makes them human.
- [The stuff between](https://betterthangood.xyz/blog/the-stuff-between/): Your service business has 'stuff zones' between every milestone. Optimizing these zones is where you become more profitable, scalable and sellable.
- [Zero busy work](https://betterthangood.xyz/blog/zero-busy-work/): AI’s value isn’t in replacing human creativity, it’s in creating space for you to do what you actually do. A simple framework for eliminating busy work in service businesses.
- [Weeknotes vol. 16: Where have all the weeknotes gone?](https://betterthangood.xyz/blog/weeknotes-16/): Back after breaking the streak. Since selling Good Work I've been juggling creating, consulting and acquiring, and making Better Than Good the home for all three.
- [Principles and systems to optimize your service business](https://betterthangood.xyz/blog/principles-and-concepts-to-optimize-your-service-business/): Ten years of running a web agency taught me what works: profit levers, finding your niche, building just enough process, and staying out of the commodity trap. Here's the practical stuff that made the difference.
- [Weeknotes vol. 15: Professional admin dashboard maker and Mac app polisher](https://betterthangood.xyz/blog/weeknotes-15/): Built a Rails admin dashboard for the production database, passed 100 beta sign-ups, and polished the Mac app's quick-add, my favorite way to capture a thought.
- [Weeknotes vol. 14: More pool, less computer](https://betterthangood.xyz/blog/weeknotes-14/): A summer week heavy on play: a few ultra-early Braindrop testers, the test suite hit 42, and I prototyped Craftman's Green, my weird new work-meets-play project.
- [Weeknotes vol. 13: May the thirtieth be with you](https://betterthangood.xyz/blog/weeknotes-13/): Down to polish before the Braindrop for Mac beta ships. It now installs as a progressive web app, I renamed all the old robotic weeknotes, and upgraded Ruby.
- [Weeknotes vol. 12: Teaser site, settings screens and new ideas](https://betterthangood.xyz/blog/weeknotes-12/): Last week of school stole my time, but we launched the Braindrop teaser site for beta sign-ups, I built the settings screen, and app number two is brewing.
- [Weeknotes vol. 11: ‘Basically done.’](https://betterthangood.xyz/blog/weeknotes-11/): Braindrop for Mac is 'basically done' and I'm using it full time. Matt's building out the test suite, and the landing page launches next week with beta sign-ups.
- [Weeknotes vol. 10: Back in the CSSaddle](https://betterthangood.xyz/blog/weeknotes-10/): Back in production CSS after years in Rails, and thanks to native variables, nesting and @import I can happily write raw CSS again. The old internet is back.
- [Weeknotes vol. 9: Micro-apps and nursery rhymes on Rails](https://betterthangood.xyz/blog/weeknotes-9/): A skip week for dad duties, but I sketched out our first micro-app, stacked up some Rails books, and built a little app to track our newborn's eat-sleep-poo.
- [Weeknotes vol. 8: Active Storage and shipping baby no. 6](https://betterthangood.xyz/blog/weeknotes-8/): Wired up file uploads with Active Storage (a breeze), shipped baby no. 6, and Phil wrapped the app design, so production HTML and CSS is up next.
- [Weeknotes vol. 7: Here ye here ye, Braindrop hath reached 100 commits](https://betterthangood.xyz/blog/weeknotes-7/): Week 7: Turned latest design into working code, can now use Braindrop to track my own thoughts. Added Web Share API to replace thought processing for MVP. Test suite in progress. 100 commits and updated Ruby!
- [Weeknotes vol. 6: Web and hybrid and native, oh my!](https://betterthangood.xyz/blog/weeknotes-6/): Week 6 at Better than good: Settled on tech stack, Mac app design progress, built API and prototyped the watchOS app.
- [Weeknotes vol. 5: A month of cool stuff, a listicle](https://betterthangood.xyz/blog/weeknotes-5/): One month into building Better than good: launched website, defined Braindrop, secured office space. April goals: marketing site, API, and working MacOS prototype.
- [Weeknotes vol. 4: Coding like it’s the 80s (with stickers)](https://betterthangood.xyz/blog/weeknotes-4/): Weeknotes: balancing simplicity with function while building Braindrop, 80s coding playlists, celebrating design wins, and limited edition coasters for early subscribers.
- [Weeknotes vol. 3: 404, Chunk not found](https://betterthangood.xyz/blog/weeknotes-3/): This week: making architectural decisions between delegated types and single table inheritance, finalizing logo and font designs that scale beautifully, and mapping out our pre/post-launch strategy for Braindrop.
- [The road to Braindrop](https://betterthangood.xyz/blog/road-to-braindrop/): After a decade building Good Work into a well-oiled machine, I’m embracing a new challenge with Better than good. Learn why I’m creating Braindrop, a simple tool to help you capture and process thoughts without interrupting your focus.
- [Weeknotes vol. 2: In the rabbit holes we find marketing goals](https://betterthangood.xyz/blog/weeknotes-2/): Weeknotes 2: On the power of accountability, the clarity that comes from designing your marketing website first, and how the creative process naturally distills down to what truly matters.
- [Weeknotes vol. 1: The end beginning of an era](https://betterthangood.xyz/blog/weeknotes-1/): Started weeknotes. New office, new website, named our first app, and beat a week-long Stripe bug that came down to a missing `window.` (*ducks*).
- [Hello, world!](https://betterthangood.xyz/blog/hello-world/): After 18 years in client services, I’m starting Better Than Good—a tiny software studio creating apps that are fast, simple and a joy to use.