The Next Input — Issue #104

In partnership with

⚡ The Briefing — 60 sec

• Google launches Gemini 3 Flash and makes it the default model in the Gemini app
  The model wars never end. If 2025 had a word of the year here at The Next Input? Cooking 🔥

• Google experiments with mini-apps inside the Gemini app via Project Opal
  That’s the old one-two. Mini apps inside the model? OpenAI took the jab—Anthropic, where you at fam?

• Volvo EX60 to be first new car in Australia with Google Gemini AI
  Might as well make it a clean three-for-three. Gemini’s not just in your phone—it’s in the driver’s seat now.

🛠️ The Playbook — The Embedded AI Experience Layer

Mission Embed AI directly into products, tools, and interfaces—so users don’t “open AI,” they just use it.
Difficulty Advanced
Build time 3–4 hours
ROI Massively increases adoption by removing friction and context switching.

0) Why This Matters

Google isn’t just shipping better models—it’s hiding them inside everything.
Apps, mini-apps, cars, assistants. No prompts. No ceremony.

That’s the real shift: AI stops being a destination and starts being infrastructure.
This playbook shows how to design for that future—now.

1) Architecture

2) Workflow

1. User performs a normal action (scrolling, driving, emailing, configuring).

2. System infers intent from context—not a typed prompt.

3. Embedded AI surfaces a suggestion or action inline.

4. User accepts, edits, or ignores—no interruption.

5. Outcome feeds back into short-term memory.

6. Over time, the AI learns when not to speak.

3) Example Prompts

Intent Inference (GPT-5-mini)

Based on this user action and context,
infer the likely intent.
If confidence is low, do nothing.
Return: intent + confidence score.

Inline Action Generator (Claude 4.5 Sonnet)

Generate a suggestion that fits naturally into the existing UI.
No explanations.
No extra options.
If it doesn’t feel obvious, don’t show it.

4) Guardrails

• Silence is better than interruption.

• Never force interaction—AI should feel optional.

• Keep outputs short and reversible.

• Log ignored suggestions as negative signals.

5) Pilot Rollout — 3 hours

1. Identify one high-frequency user flow.

2. Map natural decision points.

3. Add passive intent detection.

4. Inject one inline AI assist.

5. Measure accept vs ignore rates.

6. Iterate only on accepted moments.

6) Metrics

• Suggestion acceptance rate

• Reduction in task completion time

• Frequency of ignored interventions

• Repeat usage of embedded actions

• User satisfaction delta

Pro Tip: If users notice the AI, you’re already late. The best embeds feel invisible.

🎯 The Arsenal — Tools & Platforms

• Google Gemini APIs · Fast, embedded model access · https://ai.google.dev

• Flutter · Ship embedded AI across mobile and in-car surfaces · https://flutter.dev

• Amplitude · Measure interaction and ignore signals · https://amplitude.com

• Featurebase · Collect quiet user feedback on AI interventions · https://featurebase.app

Copy-paste prompt block:

You are an embedded AI.
Act only when confidence is high.
Prefer silence over noise.
Enhance the flow—never interrupt it.

💡 Free Office Hours

Want help implementing anything? Book a free 15-minute Office Hours slot—no sales pitch, just workflows solved.

→ https://calendly.com/aaron-cylentis/the-next-input-office-hours

AI that works like a teammate, not a chatbot

Most “AI tools” talk... a lot. Lindy actually does the work.

It builds AI agents that handle sales, marketing, support, and more.

Describe what you need, and Lindy builds it:

“Qualify sales leads”
“Summarize customer calls”
“Draft weekly reports”

The result: agents that do the busywork while your team focuses on growth.

Start now with $20 in extra credits

🕹️ Game Over

The future of AI isn’t louder—it’s closer.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: https://cylentisai.beehiiv.com/subscribe