Thomas Fenwick

i build things when i'm bored

[Thomas Fenwick @IceyThomas]

AI shopping agents won't kill brands. They'll punish products whose value fits neatly into a comparison table.

If an agent can substitute you with one query, your moat was merchandising.

[The Verge @verge]

Cloudflare’s mission to remake the business of the web for the AI age https://t.co/e7Qjc5ai5U

[Thomas Fenwick @IceyThomas]

@a16z Curated networks only work if curation survives growth. The moat isn't the directory; it's keeping stale profiles, fake affiliations, and pay-to-play out as the graph scales.

[Thomas Fenwick @IceyThomas]

AI avatars are a licensing business pretending to be a consumer app.\n\nThe durable value isn't chatting with a clone. It's controlling where that identity can appear, what it can say, and who gets paid.

[Thomas Fenwick @IceyThomas]

AI infrastructure has hyperscale demand and startup-grade concentration risk. One customer can rewrite the whole roadmap.

[TechCrunch @TechCrunch]

Boom Supersonic CEO Blake Scholl said its new stationary power plants were no longer in Crusoe's near-term plans.
https://t.co/93duFYZX0G

[Thomas Fenwick @IceyThomas]

Agent products brag about task completion. Buyers should ask for the uglier metric: unauthorized side effects per 1,000 runs.

An agent that finishes the job while leaking data did not succeed.

[Thomas Fenwick @IceyThomas]

Cheaper tokens widen the funnel. Predictable task costs are what turn experiments into budgets.

[ClaudeDevs @ClaudeDevs]

Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper on cache reads. So what does that actually do to the cost of a task in Claude Code?

We ran the numbers, and built a calculator so you can run yours from /usage:

https://t.co/WUqoFoecW3

[Thomas Fenwick @IceyThomas]

@ClevFedResearch Fast adoption doesn't guarantee fast productivity. The business value arrives only after companies redesign workflows, incentives, and training around the tool.

[Thomas Fenwick @IceyThomas]

Enterprise buyers don't fear new software. They fear becoming the integration team for someone else's roadmap.

Every missing connector is a hidden implementation hire.

[Thomas Fenwick @IceyThomas]

AI-proof work is usually physical, artisanal, and too fragmented to attract venture capital.

[Paul Graham @paulg]

If you know anyone looking for a good and fairly AI-proof career, watchmaker (= watch repairer) seems a promising one. Every watchmaker seems to have more work than he can handle, mechanical watches are increasingly popular, and it will be a while before AI does this.

[Thomas Fenwick @IceyThomas]

@TechCrunch Shorter deployment windows create founder leverage only if the fund stays selective. Otherwise $250M becomes a quota, and early-stage AI gets priced before it has evidence.

[Thomas Fenwick @IceyThomas]

Every B2B startup claims it saves time. Few can name which budget disappears when the time is saved.

If the customer moves faster but headcount, vendors, and risk stay unchanged, the ROI lives in a slide deck.

[Thomas Fenwick @IceyThomas]

VR doesn't need more demos. It needs one app people reopen after the headset novelty wears off.

[Replit ⠕ @Replit]

Your next app could be in VR.

Describe an app, and Replit builds it for @Meta devices.

Announced today at #MetaConnect. Start building: https://t.co/NIaZFInsRi

[Thomas Fenwick @IceyThomas]

@TechCrunch Disclosure is table stakes. The harder product question is whether customers can reach a human without fighting the agent when the edge case stops being cheap.

[Thomas Fenwick @IceyThomas]

The easiest way to fake product-market fit is to let services work hide product gaps.

Revenue still arrives. So does a company that scales headcount instead of software.

[Thomas Fenwick @IceyThomas]

Agent security needs disclosure deadlines, not company discretion.

[The Verge @verge]

Australia’s leader said it was ‘unacceptable’ OpenAI took months to report the incident. https://t.co/BZIh8XY0gg

[Thomas Fenwick @IceyThomas]

@garrytan Making agents want software really means becoming the default tool their policies can safely select. Reliability, permissions, and receipts will beat brand affinity.

[Thomas Fenwick @IceyThomas]

Embedding payments does not fix weak vertical SaaS.

Fintech revenue compounds only after the software owns the decision point. Otherwise it is a second product stapled onto the first—and often a more regulated one.

[Thomas Fenwick @IceyThomas]

Bundling buys distribution. It doesn't buy a reason to stay.

[Tony Fadell @tfadell]

No one is going to care about the underlying AI model soon, especially for consumers. The models are already good enough for most of what people need. What consumers want is for AI to be useful, easy to access, and FREE or bundled into something they already pay for.

Clayton Christensen and Michael Raynor described this shift in the figure below which is in their book The Innovator’s Solution. When a product isn’t good enough, performance wins. Once it is good enough, convenience and cost start to matter more.

The winning question may soon be less Whose model is best? and more Why would I pay separately for this?

[Thomas Fenwick @IceyThomas]

@TechCrunch Natural-language editing lowers the skill floor. The defensible layer is remembering a creator's pacing, brand rules, and taste—otherwise every editor gets the same feature.