Intelligence Snacks

# The Good Stuff, with Pete and Andy - Episode 37: Stocking Fillers - Predictions for 2026

Hosts: Pete and Andy (sporting festive headgear at the beach)

Pete and Andy kick off their Christmas predictions episode with zero preparation and maximum confidence. They tackle the biggest question: what will happen in 2026?

Key Moments:

[00:51] Andy's two-for-one Christmas prediction: most AI predictions won't eventuate, no AGI in 2026

[03:09] Revised timeline: feels more like a 10-year transition than 5 years

[04:02] Pete's first prediction: models become less relevant, focus shifts to agent tooling

[05:40] Are model releases becoming underwhelming? The breakthrough isn't as significant anymore

[05:53] Elon's take: "The whole framework is incorrect—throwing more data doesn't give you intelligence"

[07:08] Why software automation is so valuable: automate software, you get free option on everything else

[08:17] Less discussion about new model releases—when Opus came out, people raved for a few days then moved on

[09:05] Death of one-shot benchmarking: "Nobody uses these things in that fashion"

[10:26] The Presidio Bitcoin example: Gemini 3 got everything wrong because the task was too broad

[12:07] Flow matters: sequential tasks build on previous work, jumping around creates cognitive load

[13:15] Andy's prediction: website development will be completely commoditized

[14:01] Big agencies won't be able to justify high spends anymore—except for enterprise clients

[16:24] The WYSIWYG problem: hard to update flat file websites without developer tools

[18:13] Pete's prediction: agents move out of the terminal

[19:08] The model switching advantage: when one gets nerfed, quickly switch to another

[20:30] Why Goose is better for non-coding tasks: less sandboxed, happier to just do work

[22:06] The spare machine problem: agents need machine access to be powerful

[23:47] Most people still talk about Copilot, not Claude Code or Codex

[24:44] The Excel analogy: "This is like super Excel, why wouldn't you want to learn it?"

[27:06] Calling them "coding agents" creates mental resistance for non-coders

[27:27] The agent becomes the engineer, you become the product manager/technical architect

[30:46] The killer use case isn't software—it's the business you already have

[31:20] Service 10x the market with no additional headcount: "That's insane"

[32:29] Value will accrue to small businesses, not S&P 500 companies with cultural inertia

[33:50] Andy's prediction: AI becomes more politically negative in 2026

[34:36] Politicians will use fear-mongering based on job loss to accrue more power

[36:43] Greater divide between adopters and laggards—companies that embraced AI take big leap forward

[38:38] Small nimbler SMEs will be the standout stories, not big enterprises

[39:32] More of your business can run in software than you thought it could

[40:40] The Replit debate: "Why would I use Replit when I can just change Wingman myself?"

[41:40] Wingman needs the concept of a business inside it—metadata that flows through apps

[44:26] Pete's hot take: large models are a dead end, we don't need bigger models with more data

[45:20] "The model should understand what you're doing, not be responsible for knowing stuff"

[46:40] Fast models over big models: "I think that becomes the rallying cry"

[48:02] The inverse direction: everyone's been focusing on thinking time, slower and more deliberate

[49:12] Speed unlock: "If it took 10 seconds instead of 10 minutes, you'd use it 100 times more"

[51:15] Pete's fundamental belief: "I don't think you want the model to know stuff—it's a bug we ship as a feature"

[52:56] Why domain-specific models don't make sense: graphs do the heavy lifting of knowing

[55:05] Timeline check: specialized models and speed focus probably not a 2026 thing

[57:24] Rise of the Agents: more use, simpler to use, non-coding use cases becoming clear

[58:38] Final agreement: agents are the future, not bigger models

The Good Stuff, with Pete and Andy - Episode 36: AI Fallacies Hosts: Pete and Andy

Pete and Andy tackle common AI fallacies head-on, starting with the "junior developer" myth. They explore why juniors will actually thrive and why we're entering a golden age for small teams.

Plus: reflections from their first Touch Don't Look workshop, the death of traditional SaaS, and why Pete is "insanely excited" about where they're going with Wingman.

Key Moments:

* [01:49] The seating configuration theory: why talking side-by-side works better for blokes

* [02:10] The junior developer fallacy: "Coding AI is here, we don't need junior devs anymore"

* [03:11] The fallacy extends to junior lawyers and accountants—basically all junior roles

* [04:04] Pete's take: juniors actually adopt new tech faster because they don't have baggage

* [04:44] London law firm story: how partners explained the inefficient system

* [05:06] "The associate crosses it all out and starts again, then the senior does the same"

* [05:47] Weighing paperwork to charge: "Six inch file? That's $600,000"

* [06:25] The uncomfortable truth: junior lawyers never added value in the old system anyway* [07:30] Why couldn't a tech-savvy junior lawyer act more like a senior with better tools?

* [09:15] New business models emerge: one senior lawyer with AI could serve 1000 clients differently

* [11:00] SaaS companies are building for the average—your specific needs don't matter to them

* [14:30] The "golden age of small teams" thesis: 2-10 person teams can now compete

* [16:45] Historical precedent: juniors always adopt new technology first (mobile, cloud, etc.)

* [19:20] The real question: will there be work? Not "will juniors be employable?"

* [22:00] Why AI makes protectionism harder—you can't hide that you're not using the tools

* [24:15] People who don't adopt will look obviously incompetent compared to those who do

* [27:30] Traditional education is completely misaligned with what's needed now

* [30:45] The credentialism trap: spending $100k on degrees that don't teach relevant skills

* [33:20] "Buy a Mac Mini, get Wingman, spend a year learning—you'll be miles ahead"

* [36:15] Touch Don't Look workshop debrief: people helping each other, energy in the room

* [38:40] The realization moment: "Wait, this is on my phone? It's real?"

* [42:00] Why cohort-based learning works: people bounce ideas off each other

* [45:30] Speedrun positioning: build a CRM, website, and agent onboarding in 4 hours

* [48:15] Marginal gains model: monthly rapid prototypes for the community

* [51:20] The 1000 True Fans model: economics work when you deliver to a cohort

* [54:00] Why Nostr-based infrastructure solves authentication and authorization for free

* [56:30] "I can just give them a key, they never see it, they can sign into a thousand things"

* [58:00] Pete's excitement: "I've got big plans. Insanely excited about where this goes."

* [59:00] The education business wrapped around tech enablement with AI

* [1:01:19] Final thought: "We've landed on a nice spot"

Quote: "The fundamental fallacy is assuming that the work and the industry and the company is all packaged the same and not that there's some disruption to the business model."

# The Good Stuff, with Pete and Andy - Episode 35: The State of AI Tools

**Hosts:** Pete and Andy

**Guest:** Anthony (Dead Man Oz) - AI enthusiast, open source developer, Perth local Pete and Andy sit down with Anthony at the back of the van to discuss two and a half years of using AI coding tools. From surviving the Claude degradation period to building custom tax software, they explore multi-model planning workflows, the death of white-collar jobs, and what work looks like when kids enter the workforce.

## Key Moments:*

[01:23] Anthony's journey: two orders of magnitude improvement in AI tools over 2.5 years*

[02:14] The Claude degradation period—when the model went retarded for a month*

[04:21] The elaborate fake application: Claude invented entire interfaces that weren't wired to anything*

[08:31] Multi-model planning: Gemini says yes early, Claude next, Codex is anal retentive to the nth degree*

[14:07] Specialized sub-agents that actually work: Atomic Committer and Git rebasing tools*

[16:27] Claude as a "moany little bitch" that always wants permission*

[21:50] Corporate IT won't move quickly—they're too scared of risk assessments*

[23:45] The Excel analogy: vibe coding is the new making an Excel sheet that does a thing*

[27:33] "Previously I would have been like, what? You're going to rock up with Claude Code."*

[35:35] Anthropic study: 1,250 people, 90% find value, but 70% say there's stigma using AI*

[39:05] Protectionism: "If I admit I'm using it, can't they just replace me with AI?"*

[42:01] Pete's hot take: LLMs understand language, not facts—use databases for facts*

[47:38] Do creatives using AI tools become 100x more valuable in the short term?*

[53:04] AI conference shock: 1 in 6 submissions had fabricated references and quotes*

[54:35] Speed isn't raw speed—it's removing the lag of waiting for people and debugging cycles*

[57:09] The breakaway model myth: "Your whole premise is incorrect about escape velocity"*

[59:08] Anthony eliminated his tax agent: built custom software in two weeks*

[1:02:01] The death of SaaS: they need 10 million users, you need it to work once a year*

[1:07:19] Touch Don't Look, Speedrun, Marginal Gains—the full business model explained*

[1:10:20] Beacon project: could integrate with M-Pesa through WhatsApp, perfect for Kenya*

[1:12:09] Anthony's summer project: K-pop demon hunters crossed with Pokemon*

[1:17:09] "You can now do these things. You have more agency. You can experiment more."

The Good Stuff Episode 33: On AI and Design**Hosts:** Pete and Andy **Guest:** Jarrad Grigg, Chief Product Officer at Adapter

Pete and Andy sit down with Jarrad Grigg to explore how AI is transforming design. From Figma Make to DreamFlow, they discuss the gold rush for next-generation design tools, whether AI can be truly creative, and why craft still matters. The conversation challenges assumptions about delegation vs. doing, reveals why Excel runs enterprise, and explores how iteration speed has become the new competitive advantage.

## Key Moments:* [03:22] Jared's "oh shit" moment: train station photo with watch, GPT tells him which train to catch

* [04:46] The design tooling journey from Photoshop 6 to Figma's workflow revolution

* [10:00] The AI design tools gold rush: Figma Make, DreamFlow, and who will win

* [11:03] Hot take: design systems are mostly a waste of time (except now with AI training data)

* [13:42] Why every AI design tool starts with a text box—and why designers hate it

* [17:40] DreamFlow's approach: infinite canvas + prompting + fine-tuning controls

* [20:00] Where does the user come into the design process with these new tools?

* [23:00] Pete's sovereign engineering experience: shipping a new app every week in 5 hours

* [26:00] The $94 million Bureau of Meteorology website disaster story

* [31:00] Will AI lead to standardized, homogenous design everywhere?

* [35:00] The creativity question: AI-generated purple gradients and training data limitations

* [36:08] Pete's pipeline experiments: AI personas walking and talking through problems

* [38:44] Set and setting for AI: changing how we interact beyond the text box

* [42:37] Creativity needs time to breathe—dialogue over delegation

* [46:00] Why UI isn't dead and voice interfaces won't replace visual design

* [49:01] The typing vs. speaking debate: Andy filters ideas through writing

* [50:04] Pete's Excalidraw whiteboard workflow: red boxes, green boxes, organized chaos

* [54:04] Uncomfortable truth: loosely-coupled Excel sheets run the entire world

* [56:45] The cottage industry opportunity: building better tools for individual problems

* [1:00:10] Enterprise problems aren't technical—they're risk, compliance, and people

* [1:05:00] The craft question: does AI destroy or enable it?

* [1:08:18] The essay analogy: learning happens in ancillary exploration, not the output

* [1:10:11] Doing vs. delegating: why delegation produces mediocre results

* [1:13:40] The horse drawing meme: craft is passion and effort over time

* [1:15:01] AI gives designers more time for craft and exploration

* [1:16:32] Pete's realization: work becomes fulfilling when you own the creation process

* [1:19:53] Energy drain vs. energy gain: doing what you want with AI scaffolding

* [1:22:32] It's all about loops: iteration speed is the new competitive advantage

* [1:23:34] The quality triangle shifts when you change the underlying technology

**Quote:** "It gives us more time for craft. It's a freeing tool. If you really enjoy something it's not a job—you're constantly playing with it."

The Good Stuff, with Pete and Andy - Episode 31: Where We've Changed Our Mind on AI

00:22 - Theme: What have we changed our mind on in the last 31 weeks?01:48 - Australian Christmas: hot, beautiful, and reflective03:17 - Human at the edge model: the big shift03:34 - Why being at the edge means understanding less05:10 - Bionic human approach: involved in all decisions09:01 - Working in dialogue with AI (Gigi's approach)10:12 - Collaboration vs abdication of work10:21 - Why collaborative work is more entertaining12:15 - 10x vs 100x speed: choosing fun over max efficiency14:14 - Collaborative work as energy-giving16:02 - Permission-less collaboration: no judgment, no mood18:35 - Prompting tricks: "you're 160 IQ" and senior engineer20:02 - Writing as collaboration, not delegation21:50 - The embodied self: what AI lacks22:49 - Collaboration as the natural way to work with AI25:04 - Multi-agent systems add unnecessary complexity25:10 - Single agent with context from folders27:32 - "Do not write tests, I will test"29:04 - Diminishing returns on complexity29:53 - Test-driven development doesn't work with AI29:59 - Multi-agent systems: sounds good, doesn't work (mid-curve)30:36 - Filling gaps vs doing everything yourself39:31 - Buy vs build philosophy: the big pivot41:07 - Existing organizations are messy creatures42:31 - Timing shift: 5-year to 10-year game42:55 - Knowledge diffusion problem: nobody understands it yet45:41 - Build then buy: prove it first before acquisition46:20 - First principles vs practitioner reluctance46:43 - Paradox: successful businesses resist change most47:48 - Building capital vs burning time on acquisition48:45 - Learning by doing the job yourself50:40 - Backing operators: working in and on the business51:49 - Change management at scale: years of complexity52:03 - Gravitating back toward building from scratch53:20 - Building is more fun than acquiring53:41 - Positive vs negative energy: layoffs vs growth55:42 - Barriers to entry and regulatory hurdles56:46 - Not interested in government-regulated businesses58:37 - Small acquisition vs starting from scratch58:55 - Ownership of growth vs resistance to change01:01:28 - Why sell AI services into large companies?01:01:52 - Leverage paradox: why sell your time?01:02:53 - Nobody really understands AI implementation yet01:03:16 - Soundbite knowledge vs actual understanding01:04:06 - Wrong product: strategy requires understanding first01:05:08 - Training and education as the real market need01:05:31 - Proof of work: you have to do it01:06:01 - Speed running six months of learning

The Good Stuff, with Pete and Andy - Episode 30: AI Tools That Give Agency

00:00 - Episode 30 milestone and introduction

00:35 - Vibe coding workshop recap: tired, distracted, hectic

00:49 - Challenge: normies building Bitcoin wallets from phones

01:30 - Learning DNS, routing, and nginx on the fly

01:58 - Workshop success despite constraints

02:09 - Lost internet, crammed courtroom, Starlink saves the day

02:37 - 15 people build custom Bitcoin wallets in 30 minutes

03:33 - Group learning dynamics and organic collaboration

04:14 - Reverse engineering the Replit stack

05:12 - Why CLI tools create barriers for normies

05:55 - Inventing app hosting inside Wingman

06:22 - Building subdomain routing and DNS management

08:30 - Reverse proxying and security considerations

10:45 - Phone-based development: the ultimate accessibility test

13:00 - Wingman as "replete for your own box"

15:20 - Users own their data and infrastructure

17:30 - Local LLMs vs cloud models: the sovereignty question

20:00 - Replit's business model vs individual agency

22:45 - Building tools for non-technical users

25:15 - File browser, code editor, and hosting in one

27:30 - Workshop format: chaos, breakthrough moments, and Bitcoin transfers

30:00 - Vibe coding: removing friction from creation

32:15 - AI as enabler of individual agency

34:45 - Small business vs enterprise: different needs

37:00 - Not convincing boards, just building what works

39:30 - Corporate products vs tools for builders

42:00 - Model flexibility: switching between providers

44:15 - Data sovereignty and GitHub integration concerns

46:30 - Bringing AI into your infrastructure, not vice versa

48:45 - Local models for sensitive business data

51:00 - Model selection: right tool for the task

53:30 - Microsoft Copilot vs Wingman positioning

54:36 - Building for small business, not enterprise

56:00 - Access to models without vendor lock-in

56:55 - Enabling agency rather than creating dependency

57:20 - Data access without platform lock-in

58:13 - Model selection: cheaper models for simple tasks

59:00 - Terminal amnesia: the universal developer experience

59:15 - Future: natural language command execution

59:53 - Model lobotomization drama and platform switching

01:00:08 - "That could have been a Wingman" - wrap up

The Good Stuff - Episode 29 - From Zero to Vibe Coder

In this episode, Pete and Andy dive into the world of "vibe coding" and discuss how to get someone from absolute zero to building their own applications in record time. The conversation centers around Pete's upcoming workshop at the Bitcoin Bush Bash in Busselton, where he plans to teach 20-50 people how to create customized Bitcoin wallets from scratch in just 30 minutes.

They explore the traditional barriers to learning programming—the presupposed knowledge, the friction of setup, and the intimidating complexity—and how AI tools have dramatically changed the learning landscape. The discussion touches on the asymmetry of knowledge in tech, the challenges of teaching coding to beginners, and how AI has become the non-judgmental tutor that cuts through layers of assumed expertise.

They also explore practical applications of AI tooling beyond coding, from business automation to tax preparation, and make a compelling case for why business owners in particular need to understand these tools to stay competitive.

**Timestamps:**

- 0:00 - Introduction and the value of "always be recording"

- 2:00 - The vibe coding workshop challenge: teaching Bitcoin wallet creation in 30 minutes

- 5:00 - The friction problem: terminal commands, repo cloning, and beginner barriers

- 8:00 - Why experienced developers struggle to teach: the asymmetry of knowledge

- 11:00 - How AI cut through the learning barriers and changed everything

- 14:00 - The "Hello World" drop-off problem and learning surface area

- 18:00 - Current tools still aren't normie-friendly enough for true beginners

- 25:00 - The Replit solution: web-based coding that removes installation friction

- 30:00 - AI agents and the future of automated workflows

- 35:00 - Why business owners need to attend vibe coding workshops

- 40:00 - Moving beyond ChatGPT/Claude to agent-based tools like Wingman

- 43:00 - Using AI for non-coding tasks: tax organization and business automation

- 47:00 - The future of email-based businesses and automation opportunities

- 50:00 - Closing thoughts on freeing people up for higher-value work

Good Stuff Episode 28

Pete and Andy reunite in person at City Beach to explore how AI agents are can give individuals the power to solve their own problems. From Wingman V2's evolution to the resurrection of pre-digital work practices, they discuss why treating agents like employees with their own computers might be the key to unlocking practical AI leverage without the complexity of SaaS abstractions.

0:00 Back in the van at City Beach, Perth - discussing grass, beaches, and jet lag after Madeira

2:55 Kicking off with Wingman V2 development and the importance of reading skills documentation

5:02 What is Wingman? The TLDR for new listeners on agent orchestration software

9:05 Anthropic releases Claude Code on mobile - perfect timing for Wingman's approach

10:19 The lobotomization problem and why model flexibility matters for production systems

13:29 Building processes with agents: triggers, workflows, and file-based conventions

16:12 File watchers and convention-based programming for agent coordination

18:44 The challenge of selling Wingman vs. using it to run businesses directly

21:08 Why agents are like employees: managing workload across multiple direct reports

23:30 The cloud vs. on-premises debate: putting computers in businesses, not businesses in computers

25:42 Staying involved in the process to maintain intuition and avoid costly mistakes

27:52 The last mile problem: getting from 95% to production-ready

30:09 Small vibes, many courses: iterative development with constant testing and commits

31:38 The shift from resource allocation to rapid experimentation in enterprise

34:17 Why outsourced consulting models struggle with agent-driven development

37:20 Multi-user Wingman: the philosophical question of shared vs. individual agents

39:53 Using Nostr keys for identity management in small business tools

42:10 Building Good Stuff with just two people and AI leverage

44:45 The importance of developer logs, security reviews, and daily highlight reports

48:49 Relearning structured work practices from the pre-digital era

52:07 Building Pontefex: a visual interface bridge for Claude Code web development

54:37 Why clipboard-based workflows beat complex integrations

56:07 The Excel principle: empowering people to solve their own problems with tools