AI · Software · Business · Work
Big Ideas Compound
For people building and running modern businesses, Intelligence Snacks helps you quickly digest the big ideas emerging from the world of AI, delivered to your inbox each week.
Really enjoy your podcast every week. You guys are about a year ahead of the other talking heads on YouTube so I feel like I’m in on a secret.
Featured Snack · Privacy & Security
Responsive Video Models Could Adapt Scams
Video models that respond to a person's reactions could make digital tutors more attentive, but the same ability could also make impersonation scams more convincing.
Read the snack ↗
From the conversation
Episode 75: The Human Bottleneck
Responsive Video Models Could Adapt Scams
Video models that respond to a person's reactions could make digital tutors more attentive, but the same ability could also make impersonation scams more convincing.
AI Noise Separates Public Feeds and Group Chats
As AI-generated content spreads across] public feeds, private moderated groups give people a practical way to [decide what gets shared and who gets to participate.
A MacBook Notch App Posts to X
Andy built Sally to capture and schedule posts without opening X, where a quick publishing task can easily turn into an extended browsing session.
Disposable Artifacts for Short-Lived AI Collaboration
A temporary interactive workspace can give someone and an AI agent a place to work together without needing to save everything as a permanent document.
Match Documents to the Work Process
Documents are most useful when the work requires a lasting reference, collaborative review or a record of what happened
State Diagrams Guide Long-Running AI Agents
A visual comparison of how a system works today and how it should work after a change could help people direct AI agents working on longer tasks.
Jev Gives AI Drafts a Separate Editor
A separate review model could identify unwanted writing habits after a draft is complete, allowing the drafting model to concentrate on producing good writing rather than following an extensive list of style rules.
From the conversation
Episode 74: Normie Mode
Choose AI Models by Task and Cost
Normie Mode turns scattered model comparisons into practical recommendations for familiar jobs, balancing capability against the cost of getting the work done.
Persistent AI Assistants Can Leak Your Address
Giving an AI assistant its own computer and continuing access makes delegation more useful, but it also lets small decisions escalate into privacy and physical safety risks.
AI Agents Work Best After Human Decisions
Automation is most useful when it carries a person's decision forward, rather than deciding what they want or replacing an interaction they value.
AI Agents Need Interfaces Beyond Chat
Direct editing, precise comments and durable records keep agent actions attached to the work they concern.
AI Agents Could Replace Spreadsheets
Spreadsheets made lightweight programming accessible to almost anyone, but agents could shift some of that work into custom software with logic that's easier to inspect.
Personal AI Agents Could Increase Subscription Churn
By taking on deliberately awkward cancellations, personal AI agents could make subscription buying more episodic and expose businesses that rely on customers postponing the decision to leave.
AI Acquisitions Need Patient Operators
Buying a traditional business and automating its workflows can create value, but success depends on industry knowledge, patient change and the people who understand how the company works.
From the conversation
Episode 73: AI Model Explosion
Lower Baselines Magnify AI Model Breakthroughs
A new AI model can combine genuine improvement with a flattering comparison against an older model that seemed to deteriorate over time.
Decision Models Prioritise Mixed SEO Signals
Jev proved most useful when it ranked a steady stream of varied SEO signals for deeper analysis, rather than adding another scoring layer to a low-volume workflow.
Decision Models Keep Software Workflows Deterministic
Fast, structured model judgements can handle uncertain decision points while ordinary code remains in control of the wider automation.
Near-Free Decision Models Cut Workflow Costs
Fast, specialised models can make sorting, routing and checking cheap enough to place throughout automated workflows.
Decision Models Could Make Coding Agents Cheaper
Agent harnesses could reserve frontier models for substantive reasoning while cheaper decision models handle routine navigation, routing and tool selection.
Personal Agents Could Undercut Consumer Inertia
Persistent personal agents could weaken businesses that profit when customers postpone the work of finding a better deal.
Email Agents Can Preserve Communication Overload
Automating both ends of email can save individual effort while leaving the medium’s organisational burden intact.
Latest episode
View all episodes ↗
Episode 75 · Agents · AI Coding · Knowledge & Memory
The Human Bottleneck
Video models that respond to a person's reactions could make digital tutors more attentive, but the same ability could also make impersonation scams more convincing. As AI-generated content spreads across] public feeds, private moderated groups give people a practical way to [decide what gets shared and who gets to participate.
Explore the episode ↗Weekly Intelligence
Get next week’s
Intelligence Snacks
Email address
A handful of snacks from each week’s podcast.