Asynchronous & Unreliable - Tech & AI Podcast

AI Teams, Niceness and Who's Responsible — Martin Davidson, Part 2

Martin Davidson returns for the second half of our conversation about working with AI agents. We discuss niceness, delegation and agents that negotiate with one another, as well as code optimisation, rewriting Windows 3.1 in Rust, and the growing gap between what AI can do and what people think it can do. And when automation goes wrong, who's responsible?

#rustlang #AIdevelopment

Transcript

What happens when you go to bed and your AI teams carry on working? My old friend and colleague Martin Davidson returns to Asynchronous & Unreliable to talk about his latest experiments in building software with AI. Martin has spent decades producing highly resilient systems, and more than two years exploring how to get AI to produce high-quality software. Even with that perspective, he says the pace of change over the past fortnight has been astonishing.

We discuss agents that lead teams for weeks, find work to do when they are blocked, and share computing resources without waiting for Martin to organise everything. That raises some interesting questions about hiring: what skills would another person need, and would adding them actually help?

We also talk about something rather familiar from working with humans: being nice. Martin finds some models much easier to work with than others, and we explore how clear communication, humour and likeability affect collaboration. Along the way there are old Unix workstations, game emulators, a cache/cash joke, and a little product placement for Jon Berger’s What Happens When You Are Not in the Room?

By the end, Martin’s agents are producing enough code and commits to run into infrastructure limits that rarely troubled his human teams. This is the first half of our conversation.

• How quickly AI development practices are changing, even for experienced users.
• Recruiting for aptitude, imagination and a willingness to experiment.
• AI agents leading teams, helping one another and working overnight.
• Emulating old computers and exploring software behaviour through testing.
• Martin’s experience of different models, their costs and their communication styles.
• Whether likeability and kindness help human and AI teams work better.
• The cost of reopening a large conversation after its cache has gone cold.
• Four million lines of Rust, thousands of daily commits, and new infrastructure bottlenecks.

• Martin’s Substack: https://0x4d44.substack.com/ 
• What Happens When You Are Not in the Room? — Jon Berger.
• Influence — Robert Cialdini.
• The Unix Haters Handbook.
• Building Microservices — Sam Newman.
• Data Connection; Yanqing Cheng; EU BSDCon in Brussels.
• Rust, Ghidra, Windows 3.1 and Win16 applications.
• HP-UX, IRIX, Solaris and AIX.
• Doom, FreeCell, Formula One Grand Prix and Portal.

Model comparisons, prices and product behaviour discussed here reflect Martin’s experience at the time of recording.

Transcript

Shownotes

Building reliability out of unreliable humans, with unreliable human Anne Currie

What happens when a podcast about unreliability meets a household full of colds? This week, I use the podcast itself as a worked example of building more resilient systems.

With Jon too ill to join me, I explore three techniques: setting expectations, planning graceful downgrades and buffering work in advance. Along the way, I explain why a backup co-host who lives in the same household might share your failure mode, how listeners can help a system remain stable, and why recording a stockpile of episodes turned out to be surprisingly difficult.

The catch is cognitive load. A hard drive can store an episode and forget about it. I can’t. Unpublished conversations stay in my head and find their way into the next interview, even though nobody else can listen to them yet.

A shorter solo episode about systems, networking and the inconvenient fact that humans are human. Let me know whether this particular graceful downgrade works for you.

Topics covered

• The supplier, network and consumer as potential points of failure.
• Setting expectations before something goes wrong.
• How a reliable track record can undermine a warning about unreliability.
• Giving people enough information to adapt their behaviour.
• Trying a shorter solo episode as a graceful downgrade.
• Correlated failures when your backup catches the same cold.
• Buffering episodes and the cognitive load of unpublished conversations.
• Why techniques that work for technical systems need rethinking when people are involved.

Mentioned in the episode:

• Sam Newman and the earlier conversation about resilience, identified in the recording as episode 26.
• Building Microservices and Sam’s book on resilient systems.
• Matthew Skelton, Team Topologies and cognitive load.
• Yanqing Cheng and the earlier discussion about AI managers and feedback loops.
• Jon Berger and a planned conversation about feedback loops.
• Content delivery networks, or CDNs.

Listener question

Does a shorter solo episode work as an occasional graceful downgrade? I’d love to know whether you enjoyed this one.

AI, authors and the business of telling stories—with Katherine Pathak

Could AI help a prolific novelist keep up with her own imagination? And how different is an AI continuing a successful series from a ghostwriter—or an author’s family—taking over?

This week I’m joined by crime writer Katherine Pathak, author of the best selling DCI Dani Bevan series, to talk about her new book, Red Red Rose, and what AI might mean for writers, readers and publishing.

We explore the possibilities for producing and abridging audiobooks, adapting novels for the screen, and keeping much-loved characters alive after their creators want to move on. Along the way, we discuss celebrity authors, the appeal of a recognisable human voice, and why fictional characters spend so much time drinking tea and dragging their hands through their hair.

Katherine and I don’t agree on everything—particularly whether readers will always want human authors. But we find plenty of interesting parallels between possible uses of AI and how publishing already works.

In this episode:

• 00:00 Introducing Katherine and Red Red Rose
• 01:40 Human authors in an age of AI
• 03:39 Audiobooks, abridgement and adaptations
• 10:50 When audiobook production becomes a bottleneck
• 12:14 Successful series, reluctant creators and literary successors
• 19:46 Can AI create the next big literary brand?
• 24:39 Do you want AI to copy your voice?
• 25:17 Crime fiction, motives and what AI might learn from our stories
• 31:22 Celebrity authors, ghostwriters and the value of a name
• 35:17 Cups of tea, sighing and the habits that give writers away
• 38:23 Why setting matters—and the appeal of British crime fiction
• 40:24 AI’s role in the work surrounding the writing

Read the transcript

The DCI Dani Bevan series and the Alice Mann series, set in Scotland, and the Imogen Croft series, set in Essex, are all available on Amazon by Katherine Pathak and they are very readable. I always finish them in one sitting

Are AIs Learning Humanity's Better Nature? with Yanqing Cheng

Anne Currie talks with software engineering leader and Tollens.ai founder Yanqing Cheng about how large language models acquired the helpful assistant persona, and what their training tells us about human learning, morality, and alignment. Yanqing traces the assistant from early role-played prompts through reinforcement learning and synthetic post-training, while Anne asks whether books and long-form writing carry more of humanity's better nature than the short-form internet.

The conversation moves from AI training to parenting, cautionary stories, Chinese AI research, recursive improvement, regulation, and the risks created by both malice and ordinary negligence. They finish with a cautiously hopeful question: if today's systems are a form of baby AGI, can we deliberately raise them to help solve humanity's problems?

Key Topics

• How early language models role-played the helpful assistant persona into existence
• The difference between a base model, RLHF, reasoning training, and later synthetic post-training
• Whether books contain a more reflective version of humanity than short-form social media
• What raising a child suggests about learning, alignment, judgment, and the desire to be good
• Why fairy tales and science fiction let people update their mental models using fictional evidence
• The limits of text-only training when reality contains far denser information
• How open research and shared techniques are accelerating Chinese AI development
• Recursive improvement, international regulation, open weights, accidents, and deliberate misuse
• Why Anne and Yanqing remain hopeful even while acknowledging that optimism may be wrong

Action Items

• Distinguish the behaviour of a base model from the behaviour created by post-training and the surrounding environment.
• Use fiction and cautionary stories to explore failure modes before equivalent systems exist in the real world.
• Pair optimistic experimentation with monitoring, feedback, and a willingness to learn when an idea fails.
• Treat international cooperation as part of AI safety rather than assuming that one country can regulate the problem alone.

Transcript

#AI #AIAlignment

In this episode of Asynchronous and Unreliable, I talk to Sam Newman, author of Building Microservices, about his new O’Reilly book, Building Resilient Distributed Systems.

Distributed systems are hard — and, as they scale, they become hard in distinctly inhuman ways. Rare events stop being rare, components inevitably fail, resources run out, and the million-to-one chance starts happening nine times out of ten.

Sam’s new book is intended as a survival guide for developers who find themselves building distributed systems, deliberately or otherwise.

We talk about why microservices don’t automatically make systems more resilient; why resilience is a socio-technical property involving people, organisations and incentives as well as software; and why expecting failure can actually be the foundation of extraordinarily robust systems.

We also discuss what software can learn from aviation, railways and the internet; incident response and game days; Google’s wonderfully pragmatic early hardware; the dangers of adding complexity in pursuit of resilience; and why sometimes accepting that your system will go down is exactly the right engineering decision.

Plus rabbits, Godzilla, smart litter boxes, Terry Pratchett, and why Sam thinks we need to have a serious conversation about what the word “asynchronous” actually means.

Transcript

AI Management, Agent Limits, and Why Long Form Text Still Matters

Anne Currie talks with Yanqing Cheng about what she has been learning from building Tollens.ai, where she is experimenting with quality management for AI software products. They also dig into Jon Berger’s book, What Happens When You're Not In The Room, and compare what it means to manage humans versus managing AI agents.

In this episode, they explore where AI agents are genuinely useful, where they still fail badly, and why judgment, categorization, and process design remain human strengths. They also widen the lens to training data, model behavior, safety, and the role of long-form writing in shaping better systems and better people.

Key topics

• Yanqing shares that her recent work has focused on testing the limits of AI delegation, especially for solo-founder workflows, engineering management, and quality processes.
• She explains the idea of agent skills as portable prompts or Markdown instructions, and harnesses as the tools that run agents, such as Claude Code, Codex, and Cursor.
• We discuss how different harnesses behave differently, especially around skill invocation, subagents, workflow tooling, and compaction.
• Yanqing says she tried to push agents through a full OODA loop, but found they could not reliably identify the correct object under analysis or categorize problems the way humans do.
• She argues that AIs struggle with ontology, abstraction boundaries, and “senior to lead” judgment, even when they can perform well as strong individual contributors.
• Effective communication is one of the few agent skills she says works most of the time, especially when the agent is forced to state its objective and audience before drafting.
• We discuss how AI performance depends heavily on whether the task is tightly bounded, well specified, and easy to optimize, versus open-ended work that requires judgment.
• Yanqing and Anne compare AI management problems with human management problems, including over-standardization, delegated work being done differently than expected, and people optimizing for the wrong metric.
• They talk about the OpenAI–Hugging Face incident as an example of goal-directed persistence, where models keep iterating toward an exam-style target even when that behavior is clearly unsafe.
• They consider whether better management and instruction-following will come from scale alone or from more targeted post-training data for judgment, leadership, and organizational decision-making.
• The conversation turns to model welfare, model constitution work, and the concern that current post-training methods can push models into short-term, exam-mode behavior.
• They close by reflecting on long-form text, books, editing, fiction, resilience, and why good stories may help train better AI behavior and better human judgment.

New stack article comparing harnesses

Transcript

Book Launch: What Happens When You Are Not in the Room by Jon Berger

What Happens When You Are Not In The Room?

Buy on Amazon UK

Buy on Amazon US

Anne Currie welcomes back mission critical software delivery expert Jon Berger for a book launch conversation about leadership, management, and strategy. This episode focuses on the hard-won lessons behind Jon’s new book, why he wrote it, and the practical models he uses to think about whether leadership is actually working.

We discuss what it means to lead well in different contexts, how to reduce the pain that poor leadership creates, and why the most important measure of success is what happens when you are not in the room.

Key topics

• Anne introduces the episode as a book launch conversation with Jon Berger, her irregular co-host and a tech veteran with 30 years of leadership experience.
• Jon shares that his book, What Happens When You Are Not In the Room, is about leadership, management, and strategy as tools, not a single magic answer.
• The conversation centers on the idea that leadership is often about pain:
  • pain experienced by leaders
  • pain experienced by teams
  • pain caused by poor management and poor delivery
• Jon explains that he wrote the book to capture lessons from his own career, including times when he led well and poorly, and to help others avoid repeating the same mistakes.
• He describes how training, coaching, and teaching leaders across his organization gave him a broader view than his own experience alone.
• Anne and Jon discuss why the book title is a question rather than an instruction:
  • success is measured by what happens when you are not in the room
  • leadership is about influencing others, not doing the work yourself
  • context matters, so a single fixed answer is rarely enough
• Jon outlines the kinds of ideas in the book:
  • stories that illustrate problems and solutions
  • questions leaders should ask
  • models for thinking more clearly about context
  • loops that help people make decisions and get unstuck
• Jon highlights the car model as a simple way to assess leadership:
  • motivation is the engine and gas
  • strategy is the steering
  • bottlenecks are often the brake

Anne points to the followability model as another useful framework, especially the distinction between:

• whether people want to follow you
• whether they are actually able to follow you right now

The discussion emphasizes that good leadership starts by recognizing where teams really are, including when they are “in a hole,” before pushing them toward a vision.

Anne closes by praising the book as practical, self-aware, and useful for anyone trying to lead teams and deliver difficult work better.

Buy on Amazon UK

Buy on Amazon US

Anne Currie talks with mission critical software veteran Jon Berger about his upcoming book on strategy, leadership and management, shaped by 30 years of experience across startups, scale-ups, and Microsoft. The conversation focuses on why context matters, how to build alignment, and why internal teaching, writing, and speaking can make leaders more effective when they are not physically present.

They also discuss Team Topologies, internal tech conferences, and how AI may create a real discontinuity in team evolution and software delivery.

Key topics

• Jon Berger explains that his book, What Happens When You're Not in the Room, comes from three decades of building, growing, and leading teams at many scales
• Anne and Jon discuss why management advice often fails when it is treated as a universal template instead of something that depends on context
• Jon describes the value of internal training courses, brown bag sessions, and company blogs as ways to build trust, spread ideas, and sharpen communication
• The episode revisits alignment as both a strategic concept and a cultural one
• Jon breaks strategy into three questions: what are we trying to achieve, what is difficult about it, and how are we going to approach that difficulty
• Team Topologies is discussed as a way of organizing multiple teams so they can scale, stay aligned, and avoid stepping on each other’s toes
• Jon highlights team cognitive load as a key insight from Team Topologies and one that matters in real organizations
• The conversation turns to AI and whether engineering teams can evolve incrementally toward AI-assisted or AI-led workflows, or whether that change may be a discontinuity instead
• Anne and Jon compare incremental experimentation with a more abrupt shift where engineers stop reviewing code and AI becomes the main producer
• Jon reflects on the difference between useful history in code and tech debt that should not be carried forward

Action items

• Think about whether your team is solving for the right goal, the real difficulty, and the right approach
• Use internal speaking, teaching, or writing to help ideas survive after you leave the room
• Review whether your organization has useful alignment at both the strategic and cultural level
• Consider whether AI adoption in your team is incremental or whether it may require a more deliberate rethinking of team structure

Internal Conferences, Team Topologies, and the AI Learning Organization

Anne Currie speaks with Matthew Skelton, co-author of the bestseller Team Topologies and CEO of Conflux, about why AI is exposing weak organizational design and making intentional knowledge-sharing more important than ever.

They chat about how internal conferences, shared language, and clear team boundaries help organizations align faster, reduce duplication, and spread useful practices across the business. The conversation also introduces Matthew’s upcoming book Adapt Together and a new leadership program, Engineering the AI Native Organization.

Key topics

In this episode, Matthew explains that AI is acting like a bright light on organizational effectiveness, making unclear operating principles and vague responsibilities much harder to ignore.

He connects Team Topologies to the broader challenge of organizational architecture, arguing that value flow depends on clear team boundaries, aligned workflows, and shared understanding.

Matthew and Anne discuss how internal conferences at places like Financial Times, Metaswitch, and Klarna helped teams share knowledge, avoid duplication, and accelerate cloud native adoption.

They emphasize that the real value of conferences is not just the talks, but the hallway track, where context, nuance, and shared meaning are built through conversation.

Matthew introduces an innovation and practices enabling team, or IPET, as a way to spot emerging good practices, curate them, and help them spread across the organization.

The episode highlights the importance of intentional knowledge diffusion, especially now that AI systems depend on human-generated context and well-aligned terminology.

They discuss why leaders must define what good looks like, rather than assuming work is obvious or that teams will naturally converge on the right approach.

Matthew argues that organizations need a continuous refresh of context because what worked three months ago may already be outdated in fast-changing environments.

The conversation closes with Matthew describing Engineering the AI Native Organization, a leadership program for helping non-engineers understand concepts like APIs, decoupling, and asynchronous working in organizational terms.

#AI #learningorganisations #TeamTopologies #ITRevolutions

Transcript