AI, async, and the end of bloated teams
Summary
The AI-first software development revolution will also change how we collaborate.
Teams will become smaller.
Technologists will play multi-skilled roles.
AI-collaborators will bring back focus to deep, independent work.
The need to synchronous communication, will reduce making async-first the default work pattern.
For decades, knowledge workers have collaborated using models that haven’t served us well. My book and this website were responses to that broken collaboration model. I have no delusions of grandeur, so I imagined that change would likely be slow, but I still voiced my thoughts.
I was wrong. Change would be faster than I’d imagined. These days, “asynchronous collaboration” is a typical, even preferred interaction pattern on distributed teams, but I didn’t expect the fundamental, AI-first shift we’re experiencing today. How we build software has changed forever.
Companies that build software face an existential crisis - adapt or be irrelevant. This adaptation is revolutionary. Faced with AI-disruption, corporations must rethink team structures and move away from cumbersome hierarchies, towards leaner, more autonomous, and asynchronous working methods.
The ghosts of teams past: bloat and Brooks' law
I assume you’ve experienced the traditional software team setup: perhaps 4-6 developers, a dedicated product owner, a quality assurance specialist, a tech lead overseeing technical decisions, a designer focused purely on visuals, experimentation and research, and a project manager coordinating efforts. On paper, it looks like specialisation. In reality, it often leads to bloat.
This bloated structure leads to communication overhead. Each specialised role requires handoffs, meetings, and alignment sessions. Information gets lost in translation, feedback loops lengthen, and administrative tasks multiply. This admin overhead echoes the central theme of Fred Brooks' seminal work, "The Mythical Man-Month" – adding more people to a late software project often delays it further. Why? Because communication and coordination costs rise exponentially with team size. As dependencies increase, velocity suffers.
Adding people to a project with many iterdependencies can often delay it
AI as the great consolidator: enter the "cranes" and the "vibe"
In 2025, AI has changed the software development game forever. As I’ve written earlier, the most effective AI tools act like "cranes," capable of lifting heavy cognitive loads. These AI "cranes" are becoming adept at handling tasks that previously required distinct human roles.
Cranes can unlock new levels of productivity
While there’ll always be holdouts, I want you to consider a few examples of AI-first software development that are already mainstream.
Coding & testing: AI coding assistants like GitHub Copilot and GitLab Duo can generate code, write unit tests, and identify potential bugs. Research suggests that development is getting faster. While human oversight remains crucial, AI significantly blurs the lines between writing and testing code.
Design & frontend development: AI tools like Uizard for mockups or Framer AI for website generation can generate design mockups, translate designs into basic frontend code, or suggest UI improvements, merging aspects of the designer and frontend developer roles.
Data analysis & reporting: AI can automate data collection, analysis, and summarisation using tools like Julius AI or features within Tableau AI, reducing the need for dedicated analysts to perform routine reporting tasks.
Simpler infrastructure & deployment: Platforms like Replit Deployments and Vercel are abstracting away complex infrastructure setup and streamlining deployment processes, often with integrated AI capabilities. Smaller teams, or solo developers, can now manage the full lifecycle of their applications more easily.
End-to-end automation & "vibe coding": AI-native platforms like Replit simplify the development lifecycle from setting up a dev environment to deploying an application. While skilful developers will always be in demand, anyone can build an application today. Today, a technologist can simply “vibe” with an AI agent using natural language. While the human focuses on the spec, the agent implements the code.
The direction of travel is clear - smaller teams, with individuals who manage a broader scope of work. Yes, this shift will have a societal cost, but we’re already talking about one-person unicorns! Replit CEO, Amjad Massad, is even betting on a future where traditional coding skills become less central.
“It would be a waste of time to learn how to code… I would say learn how to think, learn how to break down problems, right? Learn how to communicate clearly, as you would with humans, but also with machines.”
Chew on that thought for a moment. Massad was a founder of CodeAcademy. The world has changed enough in a short time that he thinks learning to code could be a “waste of time”.
The future is lean, autonomous, and async-first
The writing is on the wall. Companies like Safe Superintel have a valuation of $5B with only 10 employees. Midjourney makes $4M per employee, surpassing even the TAMANNA companies. Perplexity has a valuation of $45M per employee. These are crazy numbers that show a clear trajectory.
Smaller, more capable teams: Instead of large, siloed groups, we'll see smaller teams (perhaps 3-5 people) where members possess a wider range of skills, augmented by AI. Think "T-shaped" or even "comb-shaped" individuals with deep expertise but can contribute across adjacent domains. Collaboration models like Shape-up will become even more relevant for AI-first development.
End-to-end responsibility: These smaller teams will probably own problems or features from conception to deployment. I expect to see higher ownership and fewer handoffs.
Increased individual focus: With AI handling routine tasks and potentially complex implementation details, individuals can spend more time on higher-level strategy, user needs, and focused deep work. As Massad said in a fireside chat, humans can operate at the level of intent, while machines handle the implementation. Describing the intent, however, is no trivial task.
Reduced synchronous communication: The need for constant alignment meetings decreases when teams are smaller, individuals have broader capabilities, and AI handles more integration tasks. I expect people to reserve synchronous communication for high-stakes interactions.
The changes I expect should make async-first collaboration the perfect companion to AI-first software development. We’ll work independently, demonstrate a bias for action, and lean on AI as a tireless, genius coworker. AI will also be our assistant for toilsome work — someone who doesn’t get bored or context-switched.
Need to write documentation? No problem.
Need someone to shoulder tap? Tap AI’s shoulder!
How about making sense of that design document? AI can help.
What about writing that commit message or pull request? Ask AI for help.
The promise of AI - less toil and more time for valuable work
The more we work with AI, the more asynchronous we are, by default. Fewer meetings and interruptions, more intentional communication, AI as an always-on, always-available collaborator, no bullshit jobs or managerial feudalism - what’s not to like?
The shift is coming, whether or not we like it.
I work for a consulting firm. The AI revolution presents us with both opportunities and disruption. On one hand, I expect time-and-materials billing to come under fire. If small, AI-first teams can achieve exponentially more than only a few years back, the simple model of selling hours becomes untenable. Indeed, we’re seeing a speedy shift towards outcome or value-linked pricing models.
There’s also a rising demand for higher-margin consulting expertise. While Upwork, Catalant, Talmix, Umbrex, a.Team and X-Team bring top-tier talent to the client’s doorsteps, consulting firms like my employers are also scrambling to bring such talent onboard.
Soon, work will be less about intricate coding knowledge or large numbers of billable staff. Instead, it’ll be about delivering value with AI-cranes, in small, high-performing, inclusive teams. And this change makes me feel like I was prescient when writing my book. The future will be async-first. I’m pretty sure of it now.
The era of bloated, slow-moving teams hampered by excessive communication overhead is drawing to a close. If we embrace the async-first shift, AI provides the tools to enable leaner, more inclusive and fulfilling ways of working together. The societal risks of AI-disruption notwithstanding, I’m hopeful for this leaner future.