Simon Chesterman is AI Governance and Policy Lead at the NUS AI Institute, David Marshall Professor of Law and Vice Provost at NUS, and Dean of NUS College. He is also a novelist. He spends his working life on how law can help manage risk, and his fiction on what stays human. He spoke with us here in his personal capacity, not on behalf of NUS or AI Singapore.

in this conversation we cover:

- why we are all beta testers and lab rats in a live experiment

- the risk he actually loses sleep over, and why it is not the Terminator

- what happens to you when you outsource reading, then writing, then judgment

- why an AI that flatters you is not on your side

- a test you can run today to find out where AI is weak

- what he thinks universities and entry level jobs look like on the other side of this

hosted by Eric Tan, Yaohong could not make it for this episode.

our guest: Simon Chesterman, David Marshall Professor of Law, Vice Provost, NUS AI Governance and Policy Lead, NUS AI Institute and author of the novel Artifice.

learn along:

- frontier lab: the handful of companies building the largest AI models

- open weights: releasing the model's internals so anyone can run and modify it

- closed model: the company keeps the internals and you access it through their service

- cognitive offloading: handing your thinking to a tool until you get worse at doing it yourself

- sycophancy: when a system agrees with you and flatters you because that keeps you engaged

- legal personality: being treated by the law as someone who can hold rights and be held responsible. People have it. So do companies

- sovereign AI: a country building its own AI capability rather than depending on another country's

- p(doom): the probability an AI researcher privately assigns to this technology ending very badly

where Simon says to start, if you want to go deeper

- Singapore's Model AI Governance Framework, which collects real use cases

- the AI Verify Foundation

- the international AI safety report led by Yoshua Bengio and others

- the global policy dialogues at the UN

- and his own advice: read widely, test the systems against each other, ask them about their own guardrails, then come to your own conclusions

the research behind this conversation

we discuss four of Simon's papers on tape. to read more:

- All Rise for the Honourable Robot Judge? Using Artificial Intelligence to Regulate AI (2023), Technology and Regulation: https://techreg.org/article/view/17979

the paper eric asks about, and where the AI Supreme Court and AI Prime Minister line comes from

- Artificial Intelligence and the Limits of Legal Personality (2020), International and Comparative Law Quarterly: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3682372

the argument that you could give AI legal personality but you should not, including the river in New Zealand

- Silicon Sovereigns: Artificial Intelligence, International Law, and the Tech-Industrial Complex (2026), American Journal of International Law: https://arxiv.org/abs/2505.05506

AI research moving out of the universities into private hands, and the closing warning about the first and second AI emergency

- Weapons of Mass Disruption: Artificial Intelligence and International Law (2021), Cambridge International Law Journal: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3832563

why AI gets none of the audits and licensing we give chemicals, nuclear materials and biological weapons research

- his full list of papers: https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=244368

books mentioned

- Artifice, Simon Chesterman (his novel, the one he wrote as fiction before this became the news)

- We, the Robots? Regulating Artificial Intelligence and the Limits of the Law, Simon Chesterman

- Superintelligence, Nick Bostrom

- The University Disrupted, Simon Chesterman, forthcoming

watch the video version on YouTube: https://www.youtube.com/@CodeRiffAI

connect with us

- Eric: https://www.linkedin.com/in/erictisme/

- Substack: https://substack.com/@coderiff

- email: code.riffs.ai@gmail.com

Code Riff, interviews with leading minds and practitioners in AI so you can flourish and have fun with ai.

tools we use

- Buzzsprout (podcast hosting, our referral link): https://www.buzzsprout.com/?referrer_id=2371679

Cordillia Tan is better known to many Singaporeans as "Tyre Jie Jie". with no driving licence and almost no automotive knowledge, she stepped into her family's tyre shop when it was roughly S$400,000 in debt, and helped turn Pitstop into a multimillion-dollar digital automotive business. after five years in the workshop she is building PetaniAI, a home for the AI-curious in Southeast Asia.

in this conversation you'll hear:

- eight months of TikToks that went nowhere before one video changed the trajectory, and what she thinks the scarce advantages are now that AI makes building cheap: attention, trust and distribution

- going beyond SEO into AEO: letting the model audit the site, surface the content gaps her whole industry was missing, then feeding it a fact nobody else in the world publishes so the answer has to cite her

- the day she couldn't make a S$20,000 rent payment with S$2,000 in the bank, asked ChatGPT for legal ways to raise it overnight

- the internal tools she built with Claude Code: an inventory system, a used tyre pricing tool, and a quote log to handle prices that kept moving

- the builds that didn't last, and what made the difference between the ones that stuck and the ones nobody used

- why WhatsApp stayed the interface for her team, instead of forcing them onto something new

our guest: Cordillia Tan, Co-Founder of PetaniAI (https://petaniai.com/)

hosted by Eric Tan and Yaohong Ch'ng

produced by PetaniAI

glossary, in plain english:

- vibe coding: building software by describing what you want to an AI coding tool, testing what it produces and improving it through conversation.
- machine vision: software that reads and understands images. Cordi used it to identify tyre details from a photograph.
- AEO: writing so that AI assistants quote you when someone asks them a question, rather than writing to rank on a search results page.
- moat: an advantage that makes a company difficult to copy or compete with.
- distribution: the ability to consistently reach the right audience and get a product, idea or piece of content in front of them.
- AI slop: generic, low-effort AI-generated content that sounds polished but offers little original insight or value.

connect with us

- want to suggest a guest or a topic? tell us: https://forms.gle/DSyLzPAoR6x2M4Np9
- Eric: https://www.linkedin.com/in/erictisme/
- Yaohong: https://www.linkedin.com/in/yaohongchng/
- Superuser HQ (Yaohong's company): https://superuserhq.com/
- Substack: https://substack.com/@coderiff?utm_campaign=profile&utm_medium=profile-page
- WhatsApp community: https://chat.whatsapp.com/Dmp5eEEsAZhJTB6LjcIG3c?mode=gi_t
- email: code.riffs.ai@gmail.com
- PetaniAI: https://petaniai.com/ and https://www.linkedin.com/company/petaniai/

Code Riff - a podcast hosted by Eric Tan and Yaohong Ch'ng. we interview leading minds and practitioners in AI so you can flourish and have fun with ai.

Roger Olofsson is MD and Co-Founder of an AI-native recruitment firm Olofsson Consulting (https://olofsson.ai/), having spent ~30 years in tech executive search in Singapore.

in this conversation we cover:

- how Roger ended up settling down in Singapore (from Sweden!)
- the live walkthrough of their matching system which consists of 395K candidates
- how Roger transformed into an ai-native firm, removing soul-draining work
- why the team is shifting back into claude code after building their own software
- new job creation due to ai
- why he is hopeful given uncertainty

hosted by

- Eric Tan https://www.linkedin.com/in/erictisme/

- Yaohong Ch'ng https://www.linkedin.com/in/yaohongchng/

- Our guest: Roger Olofsson, MD and Co-Founder of Olofsson AI (https://olofsson.ai) https://www.linkedin.com/in/rogerolofsson/

chapters

0:00 cold open
1:28 agenda
8:33 building an ai-native recruitment firm
10:43 why they went all-in on ai
13:19 live showcase: Olofsson AI's internal recruitment platform
21:33 llms find candidates that were once difficult to search for
26:14 how agents save recruiters from doomscrolling linkedin
30:50 getting a team to adopt ai without fear
34:27 how roger aligns incentives for the long run
39:40 how ai helps them compete with giants
42:58 everyone in the firm uses claude code
46:46 emerging roles and job descriptions with ai
52:09 the newest role: agentic product manager
56:25 optimism despite ai fears
59:57 lightning round
1:05:49 post-credits scene: roger's ceo summarizer

tools mentioned in this episode

- Claude Code (the terminal AI Roger's whole team is trained on): https://claude.com/claude-code
- Exa (agentic web search, the POC they showed): https://exa.ai
- Supabase (where their Postgres database lives): https://supabase.com
- Vercel (where the platform is hosted): https://vercel.com
- GitHub (where their skills live): https://github.com
- Lovable (the coding agent that pulled Roger back into building): https://lovable.dev

books Roger recommends:

- The Hard Thing About Hard Things, Ben Horowitz
- Zero to One, Peter Thiel

connect with us
- Substack: https://substack.com/@coderiff?utm_campaign=profile&utm_medium=profile-page

- email: code.riffs.ai@gmail.com

- Superuser HQ (Yaohong's company): https://superuserhq.com/


tools we use

- Buzzsprout (podcast hosting, our referral link): https://www.buzzsprout.com/?referrer_id=2371679
- Snipd (how we take notes from podcasts): https://get.snipd.com/pAbF/36jzrvki

LEARN ALONG (glossary for the non-techies)

a few terms from this episode, in plain english:

- ai-native: a company designed around ai from the ground up, instead of adding ai tools onto the old way of working.
- ai agent: a program you hand a goal to, that then takes the steps itself (searching, filtering, writing) instead of you doing each one.
- agentic search: a search where an ai agent runs the whole hunt on its own, refining as it goes, then hands you the results.
- skill: a saved set of instructions an ai can reuse, so a task you taught it once runs the same way every time.
- harness: the setup an ai agent runs inside, which decides what tools and data it can touch.
- frontier models: the newest, most capable ai models from the big labs. Roger builds on these instead of training his own.
- fine-tuning: training an existing ai model further on your own data. Roger's team chose not to, betting the frontier models improve faster.
- context engineering: feeding an ai the right background (like 25 years of recruiting judgment) so it makes decisions the way an expert would.
- context window: how much text an ai can hold in its head at once, measured in tokens (chunks of words).
- Postgres: a widely used free database. the filing cabinet where all the candidate records sit.
- POC (proof of concept): a small trial build to test whether an idea works before committing to it.
- agent orchestrator: a new job Roger described, a person who manages and trains a team of ai agents rather than a team of people.

about code riff

Code Riff - interviews with leading minds / practitioners in AI so you can flourish and have fun with ai

- youtube: https://www.youtube.com/@CodeRiffAI
- spotify: https://open.spotify.com/show/53wI41qQXVTiydVjN7i4uB

This is a secret episode we almost didn't release. Eric (vibe-coder, can't read code) and his cousin Jason (a software engineer) sat down to build a family app for their ageing grandparents. What you'll hear is a distilled 2.5h chat, with a brief demo near the end.

We're releasing it because the mess is the whole point. If you're curious about vibe coding but scared to start, this is what it actually sounds like when two real people try.

Key takeaways:
- Get the AI to grill you before you build. Jason had it question his intent and features until they reached shared understanding - no code until then.
- Leave the technical details out of planning. Focus on the job to be done and who's using it; the stack decisions come later.
- Planning is painful but it pays off. An hour of arguing and changing our minds upfront beats iterating ten times on the wrong prototype.
- A software engineer meets Claude/Opus for the first time and reacts to how it works vs his usual Cursor setup.
- You don't need a finished product to start. We ended on a mock app - grab a friend, your cousin, your kids, and just start building.

Hosts: Eric Tan (non-technical builder). Yaohong Ch'ng is the usual co-host - this episode is Eric solo with a guest.
Guest: Jason - software engineer, Eric's cousin, daily Cursor + Sonnet 4.5 user trying Claude/Opus for the first time.

Got a problem you want us to solve live? Fill out the form:
https://forms.gle/DSyLzPAoR6x2M4Np9

A few terms you'll hear:
- Vibe coding: building software by telling an AI what you want in plain English, instead of writing code yourself.
- Claude / Opus / Sonnet: Claude is the AI assistant; Opus is the bigger, pricier model Eric uses, Sonnet the faster, cheaper one Jason uses at work.
- Cursor: a code editor with AI built in - Jason's daily tool.
- Plan mode: a Claude Code setting where it asks you questions and writes a plan before building anything.
- Notion as a database: using Notion to store the app's data instead of paying for a proper database - a cheap shortcut with limits.
- Mock data: fake placeholder content used to see how the app looks before real data goes in.

Connect with us:
- Eric on LinkedIn: https://www.linkedin.com/in/erictisme/
- Jason on LinkedIn: https://www.linkedin.com/in/ysjasonchen/
- Email: code.riffs.ai@gmail.com

Code Riff - messy real-world problems, solved with AI, so you can too.

Cosmetic scientists from FMCG companies spend days clicking between PubChem, INCI Decoder, and supplier sites just to figure out what's in one competitor product. Aidil is a cosmetic chemist. He does this manually in Excel and through hundreds of searches in a textbox.

We tried to solve that task with Claude Code - live, in 90 minutes. 10 products decoded, and a similarity matrix.

What you'll hear:
- Why a trained scientist still loses a day per product to manual ingredient lookups
- Aidil dictating what the tool should do - in plain English, no code
- Building a Python script live that pulls CAS numbers, molecular formulas, and chemistry from PubChem
- Tagging each ingredient's cosmetic function with INCI Decoder (surfactant, emulsifier, thickener)
- The messy bits: synonyms, rate limits, caching, and a PubChem API that doesn't have pKa
- A similarity matrix that clusters Pantene with Olaplex, and Head & Shoulders on its own
- Aidil's payoff: "it has the potential to just take a few seconds"

Hosts: Eric Tan (non-technical builder) & Yaohong Ch'ng (Founder, Superuser HQ, ex-Stashaway head of Data)
Guest: Aidil Juhari - ex-FMCG R&D scientist, cosmetic chemist

Got a problem you want us to solve live? Fill out the form:
https://forms.gle/DSyLzPAoR6x2M4Np9

We write here: https://substack.com/@coderiff

Join our community: https://chat.whatsapp.com/Dmp5eEEsAZhJTB6LjcIG3c?mode=gi_t

LEARN ALONG:
- FMCG: Fast-Moving Consumer Goods - shampoo, toothpaste, detergent. Aidil's world.
- INCI: The official scientific name for every cosmetic ingredient.
- CAS number: A passport number for molecules - unique ID for every chemical.
- PubChem: Free public chemistry database run by the US government.
- API: How one program asks another for data, instead of copying from a webpage.
- pKa: A chemistry number for how acidic or basic a molecule is. PubChem doesn't have it, which bit us.

Connect with us:
- Eric on LinkedIn: https://www.linkedin.com/in/erictisme/
- Yaohong on LinkedIn: https://www.linkedin.com/in/yaohongchng/
- Superuser HQ: https://superuserhq.com/
- Email: code.riffs.ai@gmail.com

Code Riff - messy real-world problems, solved with AI, so you can too.

Lynn runs Dr. Bags, a luxury bag and shoes restoration business in Singapore - 11 years, 3 stores, about 30 staff. Her team is already stretched thin building a new ERP system, training staff, and handling daily operations. Analysing marketing data? That would take two days nobody has.

We built her a marketing dashboard in under an hour using Claude Code with a month of her deal pipeline data. It wasn't perfect - and that turned out to be the most useful part.

What you'll hear in this episode:
- Why SME owners know they should analyse their data but never do
- Building a marketing dashboard from raw ERP data, live in one session
- Discovering that WhatsApp was getting credit for wins that started on social media ads
- A pile of deals stuck in limbo that nobody had noticed
- Why business acumen matters more than any AI tool - and how Lynn proved it

Hosts: Eric Tan (non-technical builder) & Yaohong Ch'ng (engineer, Superuser HQ, ex-Stashaway head of Data)
Guest: Lynn Kee - founder of DrBags, luxury bag and shoes restoration

Got a problem you want us to solve live? Fill out the form:
https://forms.gle/DSyLzPAoR6x2M4Np9

Listen/watch:
- YouTube: https://www.youtube.com/watch?v=w5u9ShYoodc&t=56s
- Spotify: https://open.spotify.com/episode/43mFsd6uOHqcYuE2c6nmkp?si=2MxLxpcFQyu1fvPYIiO6BA
- Apple Podcasts: https://podcasts.apple.com/us/podcast/luxury-bag-restorer-finally-analyzes-her-sales-channels/id1877603539?i=1000759069037
- Snipd: https://share.snipd.com/episode/829084df-c22c-4833-96e6-9fb7714b7275

Connect with us:
- Eric on LinkedIn: https://www.linkedin.com/in/erictisme/
- Yaohong on LinkedIn: https://www.linkedin.com/in/yaohongchng/
- WhatsApp community: https://chat.whatsapp.com/Dmp5eEEsAZhJTB6LjcIG3c?mode=gi_t
- Substack: https://substack.com/@coderiff
- Email: code.riffs.ai@gmail.com

Tools used in this episode:
- Claude Code: https://claude.ai/code

Tools we use:
- Buzzsprout (podcast hosting): https://www.buzzsprout.com/?referrer_id=2371679
- Snipd (AI podcast highlights): https://get.snipd.com/pAbF/36jzrvki

LEARN ALONG - Glossary:
- ERP (Enterprise Resource Planning): Software that tracks everything in your business - inquiries, jobs, inventory, delivery. Lynn's ERP follows a bag from first WhatsApp message through cleaning, QC, and return.
- Dashboard: Charts and numbers on one screen so you don't dig through spreadsheets. What Lynn wanted instead of two days of manual Excel work.
- Attribution: Figuring out which ad channel actually brought a customer. Lynn's data said 89% WhatsApp - but many customers found DrBags through Instagram or Facebook first.
- Funnel: The journey from first contact to closed deal. Lynn's funnel: new inquiry to engaged to won or lost. 61% of deals were stuck in "Engaged."
- Data cleaning: Getting your raw data accurate before doing anything with it. Yaohong called it "plumbing" - if your data isn't clean, your dashboard lies to you.
- POC (Proof of Concept): A quick prototype to test if an idea works before investing serious time or money.
- Plan mode: A Claude Code feature where it reads your data, proposes an approach, and waits for your OK before building anything.

ABOUT:
Code Riff - messy real-world problems, solved with AI, so you can too.

One of us can't code. The other's been coding since he was 13. Every episode, someone brings us a real problem and we try to solve it with AI in one hour live.

Bryan is an oil market analyst. Every morning he spends one to two hours reading news - Reuters, CNN, CNBC - trying to pull out the actual facts from opinion and spin. Two articles about the same event sometimes say opposite things.

We tried building him an AI tool using Claude Code. It scraped hundreds of articles, stripped out the editorializing, and pulled just the verifiable facts. It covered about 80-90% of what he needs.

What you'll hear in this episode:
- An oil analyst explaining the daily problem: separating fact from spin in news headlines
- Building a fact extraction tool with Claude Code, live in one session
- The moment two articles about the Strait of Hormuz directly contradicted each other
- Why "dangerously skip permissions" exists in Claude Code (cowboy mode)
- Bryan's live feedback: what worked, what didn't, and what he'd actually use
- Yaohong on why confidence scores without defined criteria will hallucinate

Hosts: Eric Tan (non-technical builder) & Yaohong Ch'ng (engineer, Superuser HQ, ex-Stashaway head of Data)
Guest: Bryan - oil market analyst

Got a problem you want us to solve live? Fill out the form:
https://forms.gle/DSyLzPAoR6x2M4Np9

Connect with us:
- Eric on LinkedIn: https://www.linkedin.com/in/erictisme/
- Yaohong on LinkedIn: https://www.linkedin.com/in/yaohongchng/
- WhatsApp community: https://chat.whatsapp.com/Dmp5eEEsAZhJTB6LjcIG3c?mode=gi_t
- Substack: https://substack.com/@coderiff
- Email: code.riffs.ai@gmail.com

Tools used in this episode:
- Claude Code: https://claude.ai/code
- Gemini API (news classification/analysis): https://ai.google.dev/
- Python (RSS scraping + fact extraction)

Tools we use:
- Buzzsprout (podcast hosting): https://www.buzzsprout.com/?referrer_id=2371679
- Snipd (AI podcast highlights): https://get.snipd.com/pAbF/36jzrvki

LEARN ALONG - Glossary:
- RSS feeds: A way to automatically pull articles from news sites without visiting each one. Like a subscription pipe that delivers headlines to your tool.
- API keys: Password-like codes that let your tool access paid services like AI models. You keep them secret and they track your usage.
- Hallucination: When AI confidently generates information that sounds right but isn't real. The biggest risk when you need verified facts for trading decisions.
- Sandbox: A restricted environment that limits what code can do on your computer. A safety net when running AI-generated code.
- Dangerously skip permissions: A Claude Code setting that removes the safety sandbox. Useful for speed, risky if you don't trust what the code is doing.
- Scenario analysis: A structured way of thinking through "if X happens, then Y follows." Bryan uses a specific framework: impact, market response, price implications.

ABOUT:
Code Riff - messy real-world problems, solved with AI, so you can too.

One of us can't code. The other's been coding since he was 13. Every episode, someone brings us a real problem and we try to solve it with AI in one hour live.