The Likely Times

AvoidBrowser-automation wrappers and "AI that clicks buttons for you" startups get absorbed into the platform. If your moat is the automation layer itself, it's a feature by Q4.

Window openDivided government = legislative gridlock through 2028. Rules will come from agencies and courts instead — watch FTC and state AGs, not Congress. Regulatory arbitrage stays viable longer.

Window openExpensive money persists: plan runway and pricing for high-rate conditions, not the cut cycle. Any surprise easing is upside to bank, not a base case to build on.

Window openLow but not ignorable — and resolving in months, not years. Products sold as cost-cutting survive both branches; products sold as growth spend only survive one. Position the pitch now, not after the data turns.

Window openThe cheap-mass-to-orbit unlock is delayed, not dead. Every quarter of slip extends the window before space-hardware startups can be funded on unit economics — room to position for the cost curve before it arrives.

Build on itFrontier-grade coding at self-host prices. Compliance-heavy niches (defense, health, banking) that couldn't send code to an API become addressable. Margins on "AI coding for regulated industries" open up.

Build on itVoice becomes a commodity input, like a keyboard. The opportunity moves up-stack: voice-native vertical apps (sales coaching, language tutoring, elder care) where the workflow is the product, not the voice.

AvoidOral parity collapses the injection-logistics moat: cold-chain delivery, injection-support telehealth, and adherence apps built around needles lose their reason to exist. Compounding-pharmacy plays get hit hardest.

Window openUntil this lands, human-in-the-loop orchestration is the product. A 12–18 month window for tools that make agent supervision cheap — and a pivot deadline for when it closes.

Build on itA mega-IPO resets AI comps upward and unlocks a wave of liquid angel money from lab employees — historically the trigger for a new startup cohort (the PayPal→2003, Uber→2019 pattern). Expect the seed market to heat 6–12 months after.

Build on itProducts that are unit-economics-negative today (AI on every email, every document, every frame of video) become viable. Plan pricing for where costs will be, not where they are.

Window openAn inverted prediction: the gap that stays open. Memory infrastructure, user-context layers, and personalization stores remain third-party territory — one of the few picks-and-shovels plays the labs keep not filling.

Window openThe bottleneck it exposes isn't the AI — it's verification: replication, lab automation, and review infrastructure. Tools that let humans audit machine-generated science are the picks-and-shovels play, and nobody owns that yet.

Build on itWhoever the nominee, the 2027–28 cycle is a spend surge: campaign tech, political ads, voter data. But note this market resolves the *California-model* question — a Newsom win nationalizes CA-style tech regulation as the Democratic platform.

Window openNo single 2028 outcome trades above ~40% — the honest read is deep uncertainty. Don't build anything whose viability depends on one administration's policy; two-year regulatory horizons max.

AvoidDon't build compliance products around any single state's AI law (CO, CA, TX) as the permanent regime — a 44% chance the whole map gets redrawn federally. Multi-framework compliance tooling is the safer bet than single-state depth.