Most teams, intentionally or by default, digitally record their knowledge along a single axis - usually time (slack, emails etc). Real knowledge leverage (human and agentic!) comes from using all three axes deliberately and connecting information across them.

Here’s how I’ve come to think about it (my ontology for this problem space, for those who like it meta)…

The problem is structural, not technical

Why is it so hard to find things in Slack? Why does Notion inevitably bloat into a graveyard of half-maintained pages?

It’s tempting to blame the tools. But most of the time, the real issue is that we’re collapsing fundamentally different kinds of information into a single organising strategy - usually whatever the tool defaults to.

Information that is organised by content and cross-linked into a wider web becomes easier to understand and more discoverable. Other people - or agents - can refine it, contribute to it, or link to it from elsewhere. It grows and enriches over time.

The key insight is this: there are three natural axes for organising knowledge - time, truth, and ontology - and most teams collapse everything onto just one of them. The power isn’t in choosing between the three. It’s in using all three deliberately, for the right kinds of information, and then connecting across them.

A decision log (time) should link to the architectural principle it invoked (ontology) and update the system configuration it changed (truth). A team roster (truth) should reference the meeting where the restructure was agreed (time) and connect to the domain concepts that team owns (ontology). When these axes work in isolation, each one eventually fails in predictable ways. When they’re connected, they reinforce each other and the whole system becomes more than the sum of its parts.

The three axes (and how each one fails alone)

Each axis has a natural domain where it’s the right primary organiser. But each also has a characteristic failure mode - one that is largely solved by connecting to the other two.

1. Point in TIME

Use this when it matters when something happened.

Meeting notes. Chat transcripts. Decision logs. Incident timelines. Who said what, and when.

2. Current TRUTH

Use this when the content represents the current state of something.

People profiles. Team compositions. System configurations. Active contracts. Architectural decisions in force.

3. Knowledge ONTOLOGY

Use this when information needs to be connected into a rich web of related concepts.

Glossaries. Domain models. “What is a PSR?” type questions. Architectural principles. Design patterns and their relationships to each other.

A common anti-pattern: folder/file thinking

There’s a fourth pattern that deserves attention, not because it’s a good organising axis, but because it’s the one many people have been pattern-entrained into thinking with: hierarchical folder structures.

Folders feel intuitive because they mirror physical filing cabinets. But they impose a rigid ontology (!) that breaks down almost immediately in practice. Where does a document about a client’s technical architecture go-under the client folder, the architecture folder, or the project folder? The moment you have to choose one location for something that belongs in multiple contexts, you’ve created a discoverability problem.

Folder structures force single-inheritance thinking onto information that is inherently multi-dimensional. They work for personal file storage where one person’s mental model is the only one that matters. They fail at organisational scale, where multiple people with different mental models need to find the same things.

The pathology is predictable: people create shortcuts, duplicates, and “see also” files. The structure becomes a fiction maintained by convention rather than utility. Eventually, everyone just uses search-which means the folder hierarchy was doing nothing useful in the first place.

Why this matters more in the age of agents

If you’ve been following my recent thinking on agentic AI and organisational design, you’ll recognise why this connected, multi-axis approach matters more than ever.

Autonomous agents need to find, understand, and act on information. They can’t tap a colleague on the shoulder and ask “where did we put that decision about the API migration?” They need information to be discoverable without ambient human context, unambiguous about whether it represents current truth or historical record, and connected to related concepts so they can reason about implications.

Here’s the critical point: an agent’s ability to reason effectively is directly proportional to how well your knowledge axes are connected. An agent operating against a time-organised Slack history will drown in noise. An agent operating against isolated truth pages won’t understand why things are the way they are. But an agent operating against a connected system-where truth records link to their decision history and anchor into a domain ontology-can trace reasoning chains, verify currency, and make contextual decisions.

This means your knowledge management strategy isn’t just an internal productivity concern anymore. It’s becoming infrastructure for autonomous work. The organisations that build well-connected knowledge systems will have agents that can operate with meaningful autonomy. Those that dump everything into a single time-ordered stream will find their agents are only as good as the disorganised knowledge they’re built on top of.

The connective tissue

Understanding the three axes individually is necessary but not sufficient. The real leverage comes from the links between them. Think of it as a triangle where each edge represents a different kind of connection:

• Time → Truth: Every time-based event that changes the current state of something should update-or at least link to-the relevant truth record. “We decided in Tuesday’s architecture review to deprecate Service X” should result in the Service X truth page reflecting that deprecation, with a link back to the decision.

• Truth → Ontology: Every truth record should be anchored in the ontology. A team roster entry doesn’t just say who’s on the team-it links to the domain concepts that team owns. A system configuration page links to the architectural patterns it implements.

• Ontology → Time: Concepts in the ontology should link to their history. When did we adopt this pattern? What decisions led to this principle? This gives newcomers (human or agent) the ability to understand not just what the current knowledge is, but why it’s that way.

When all three edges of this triangle are maintained, you get something qualitatively different from any single axis alone: a knowledge system that is simultaneously navigable, trustworthy, and historically grounded. Each piece of information has context, currency, and connections. That’s the difference between a knowledge base and a knowledge system.

Practical heuristics

Each piece of information should have a primary axis-the one that determines where it lives and how it’s structured:

Primary axis is time when the sequence matters more than the current state. Audit trails. Communication records. Incident response logs. Decision journals.

Primary axis is truth when people (or agents) will ask “what is X right now?” Team rosters. System ownership. Active architectural decisions. Configuration.

Primary axis is ontology when the relationships between concepts matter more than any single fact. Domain knowledge. Organisational playbooks. Design principles and their trade-offs.

But regardless of which axis is primary, always ask: what should this link to on the other two axes? That question-more than any tool selection or taxonomy design-is what separates knowledge management that compounds from knowledge management that decays.

What to do next time

The next time you’re setting up a new Notion workspace, choosing how to structure your Confluence, or designing the knowledge layer that your agent fleet will operate against - don’t just ask “what is the primary organising axis here?” Ask the harder question: how will information on this axis connect to the other two?

The tools don’t matter nearly as much as the strategy behind how you use them (see Spine Model!). A Notion page with the right links is worth more than a perfectly categorised folder tree. A Slack thread that updates a truth record and references a concept is worth more than a thousand unlinked messages.

Information architecture isn’t glamorous work. But in a world where both humans and agents need to find, understand, and act on shared knowledge, the organisations that connect their knowledge across all three axes - time, truth, and ontology - will find that their information compounds rather than decays. And that might be the most leveraged investment you can make.

AI is driving the marginal cost of near real time digital generation toward zero.

Words. Images. Videos. Plans. Forecasts. Strategies. Entire operating narratives.

But reality does not follow the same curve.

• Land is finite.
• Energy requires infrastructure.
• Machines wear out.
• People get tired.
• Time does not compress.

We are entering a structural divergence:

Symbolic abundance. Embodied constraint.

The Real Divide

The future is not online vs offline.

It is symbolic systems vs embodied systems.

Symbolic:

• Models
• Dashboards
• Policies
• AI-generated reasoning

Embodied:

• Energy
• Materials
• Skilled labour
• Physical risk
• Maintenance

Symbols now scale at near-zero cost.

Embodiment does not.

Decision-Making in Simulation

AI can generate plausible worlds.

Market scenarios. Organisational designs. Technical architectures. Policy arguments.

Decisions will increasingly be made inside models before touching reality.

This will feel efficient. It will also increase fragility.

When plans are cheap, overconfidence is cheap.

A coherent model is not proof.

A persuasive argument is not resilience.

If it can be generated, it is not evidence.

Evidence requires contact with the physical world.

Scarcity Shifts

As symbols become abundant, scarcity moves to:

• Attention
• Trust
• Grounded measurement

Reality becomes a premium signal.

Organisations will split into two loops:

• Fast symbolic loop – generate, simulate, propose.
• Slow embodied loop – execute, measure, correct.

The advantage goes to those who close the loop tightly:

claim -> test -> measure -> adapt.

The Prescient View

AI will make simulation abundant. Reality will remain constrained by matter and energy.

The widening gap between them is where systems fail - or where durable advantage is built.

The task is not to resist hyperreality.

It is to operate within it without forgetting the territory.

“AI Agents” are becoming the basic building blocks with which to compose agentic solutions.

Most people’s introduction to AI was through ChatGPT, and most early experiments were “chatbots”. From what I can see at the moment, much of the thinking today is now stuck in one of a few narrow paradigms: they search, they follow a workflow, they answer in chat.

Useful, but shallow.

The interesting question is what happens when we step back, map the space of agent paradigms, combine them systematically, and then strange-loop them back on themselves to generate deeper behaviour?

This is about that.

The Core Paradigms of AI Agents

My working definition of AI agent for this articles purposes is: An AI agent is a digital system that can perceive digital inputs, reason about them, and take actions (via language, tools, or workflows) toward achieving a goal with some degree of autonomy. Its behaviour is largely defined declaratively, not procedurally, often in plain human language.

In my current view and at the highest level, individual agent behavior falls into four categories:

• Reactive: respond to queries or stimuli
• Procedural: follow structured sequences or tool protocols
• Autonomous: pursue goals, plan, and adapt
• Generative: synthesise novel outputs

From this, we can classify the common paradigms:

Mapping to Enterprise Contexts

Enterprises can deploy these paradigms in recognisable patterns:

These core paradigms are reasoning about agents as individual atomic entities, which is a good foundation. The real leverage emerges when these paradigms combine, and the agents are looped back onto themselves to allow emergent behaviours to…emerge…

Patterns for Combining Agents

Strange Loops: From Tasks to Emergence

Combining paradigms is powerful. But deeper intelligence comes from strange loops - agents turning back on themselves in reflective cycles.

Douglas Hofstadter popularised the term strange loop to describe systems that turn back on themselves in ways that produce emergent complexity - consciousness, self, music, mathematics. The question for AI today: can we engineer strange loops into agent architectures so they don’t just do tasks, but begin to exhibit deeper, more general intelligence?

The key unlock as I see it: each loop must introduce new evidence, structure, or optimisation pressure. Otherwise it is just an echo chamber.

Mechanisms That Might Work

This is all just speculation from my version of first principles, with a bit of GPT battle testing - if you disagree or are seeing worked exampled of these let’s talk (please!):

• Planner -> Actor -> Critic -> Replanner: externalised evaluation drives learning
• Self-ask, self-debate: role rotation prevents monoculture collapse
• Evaluator-Optimiser loops: evolutionary or Bayesian search over tool and action space
• Predictor -> Actor -> Measurer -> Update cycles: Create world-model definitions with agentic loops, keeping agents grounded with cause-effect feedback (Thanks Andries for this one!)
• Curriculum growth: agents generate harder counter-tasks for one another
• Tool-augmented verification: schema checks, linters, policy engines prevent drift
• Compression-driven progress: loops only count if they improve abstraction or accuracy

Failure Modes

Strange loops can easily go wrong if not designed carefully. The most common failure modes are:

• Limit cycles: the agent keeps circling around the same idea, just rephrasing it instead of making real progress. It looks like movement, but nothing new is added.
• Mode collapse: the system latches onto the first reasonable explanation or plan and never explores alternatives. This kills creativity and robustness.
• Reward hacking: if you give the agents a scoring function, they will find ways to game the score instead of solving the actual problem (classic Goodhart’s Law).
• Retrieval echo: the agents pull in their own past outputs as if they were fresh evidence, reinforcing earlier mistakes and creating a self-referential loop with no emergent benefits.

Each of these could be mitigated with a few practical safeguards: inject novelty at every iteration (new data, new perspectives), use multiple evaluation metrics instead of a single score, track provenance to avoid recycling outputs, and apply external checks (tests, validators, policies) to ground the loop in reality.

The last one is for me the strongest: Using this approach to solve real world problems gives an opportunity for real world anchors & fitness functions.

A Practical Template: “RAVL”

Minimal strange loop structure:

• [R]eflect: reframe, hypothesise
• [A]ct: tool calls, state updates
• [V]erify: apply hard checks
• [L]earn: update assumptions and abstractions

Stop when the delta falls below threshold or a resource budget runs out (tokens/etc).

Why This Matters

Straight-line agents - input, maybe a tool call, output - are useful but brittle.

Agentic systems, particularly ones constructed into strange loops and engineered with grounding and novelty, open the door to emergent organisational intelligence.

To be very clear: Not consciousness, not magic, but systems that refine their own framing, increase their own competence, and compress knowledge into abstractions we cannot see unaided or in advance, but can be directed for useful outcomes.

Conclusion & Next Areas to Explore

I’ve been searching for AI paradigm shift that takes us from faster horses to cars (a useful if rather tired metaphor). Getting multi agentic orchestration right could be the true leap from automation to digital intelligence.

When it comes to constraining and guiding these systems effectively, I have another idea incubating on how to include agent constitutions with executable specifications using the Spine Model, and hopefully will have some working code to share on this soon.

Another realisation that has come from doing this work is how important it is to make a distinction between “digital intelligence” and “analog/real world intelligence”. I think it might be a very important distinction, and will be the subject of an upcoming post if I can put my finger on how to explain the concepts a bit more clearly and they survive some analog intelligence tests with friends.

If this has been useful to you, please let me know how - it gives me fuel for writing more of these ideas down.

After 25+ years in software engineering and leading global teams across platforms and industries, I’ve seen multiple waves of technological change. But the emergence of agentic AI—systems that act, not just assist—represents a shift more profound than anything prior. It’s not just an evolution in tooling. It’s a transformation in how software is created, operated, and evolved.

This post is a checkpoint for myself—and a guide for others who’ve spent years building mature, human-centered engineering teams. If you already know how to lead teams that deliver, now’s the time to recalibrate. The ground is shifting quickly.

We need you.

What’s Changing?

If we think of agentic tools merely as faster ways to write software, we’re falling into “faster horses” thinking. The shift isn’t just about building the same systems faster—it’s about different forms of systems, made and operated in fundamentally new ways.

The rise of agentic AI tooling is more than a wave of productivity hacks or code co-pilots. It’s a fundamental shift in how we conceptualise software creation, collaboration, and execution. And if you haven’t been hands-on with these tools yet, you’re not just missing out—you may be planning for a world that’s already disappearing.

Agentic systems don’t just assist—they act. They operate across tools, interpret context, pursue goals, and make decisions—sometimes intelligent, sometimes deeply flawed. Crucially, these agents run in the real world, not just in test environments or behind review gates. They update data, trigger deployments, reconfigure infrastructure, and take domain-specific actions that no one explicitly coded line-by-line.

So if you’re still thinking in terms of “tools that support engineers,” it’s time to shift to “systems that are engineers.” That requires a recalibration of long-held beliefs.

Start with Your Existing Mental Models

If you’re like me, you’ve developed strong intuitions about what makes good software and effective teams. These aren’t wrong—they’re just incomplete for the new context we’re entering. The principles of clean code, clear architecture, and strong team dynamics remain crucial, but they need to be viewed through a new lens.

Mental Models & Principles to Recalibrate

1. Expertise as Bottleneck → Context as Substrate

We’ve long prized deep human expertise as the cornerstone of quality. But agents thrive not on knowing more, but on being given the right context. Your new job: shaping that context. Standardising interfaces. Curating knowledge. Designing work in ways that make it legible to machines.

2. Architecture as Fixed Plan → Architecture as Dialogue

In the agentic paradigm, architecture emerges iteratively. Agents can critique, revise, and prototype structures continuously. The architecture function becomes less about upfront definition and more about constraint management and pattern evolution over time—often with agents, not just for them.

3. Developer Productivity → System Throughput

We’ve measured impact by individual productivity, or for those thinking systemically, team throughput. Now, the key metric is system throughput: how effectively your team and its agents can solve problems end-to-end. We’ll need to rewire how we think about velocity, contribution, and team design.

4. Code as Conversation

Traditional development involved writing code line-by-line, with reviews focusing on implementation details. With AI tools, code becomes more conversational—you describe intent and outcomes, and the AI proposes implementations. This shifts the focus from “how to write it” to “how to describe what we want”, and continually monitor for drift in the real world.

5. Velocity vs. Understanding

AI tools can generate code rapidly, but this speed can mask comprehension gaps. Your role becomes less about reviewing syntax and more about ensuring your team:

• Understands the system’s architecture and boundaries
• Can effectively communicate requirements to AI tools
• Maintains a clear mental model of both the codebase and the domain

6. From Specialist to Synthesist

Deep technical specialisation is still valuable—but increasingly needs to be paired with:

• Strong system design principles
• Effective prompt engineering skills
• Ability to validate and integrate AI-generated solutions while they are running

Where to Start If You’re Already Playing with AI

If you’re already familiar with ChatGPT or using code copilots, it’s time to move beyond enhanced software construction and look at how the systems themselves will be designed differently in an agentic world.

• Use the Tools Personally
  Install GPT-based agents locally or in agent-enabled IDEs. Try multi-agent frameworks. Give yourself real tasks to solve (repeat them kata-style). Don’t outsource the learning—experience the new ergonomics of agency.

• Codify What You Know
  The best agents are trained on your ways of working. Document your playbooks, architectural heuristics, and domain insights as structured prompts or decision trees. You’re not just enabling people—you’re enabling your future AI collaborators.

• Rethink the “Team” Interface
  Think beyond roles. Rethink ownership, observability, and accountability in systems where decision-making is increasingly non-human in origin.

• Model the Future
  Sketch the org you would build if agents could replace 40% of your engineering effort. How would onboarding change? How would you pair? What would code reviews look like? This isn’t about cost cutting—it’s about adaptation.

How to Start From Scratch

If you’re completely new to agentic tools:

1. Start Small
   • Use AI for documentation, test case generation, or bug explanations
   • Ask AI to explain existing code before generating new code
   • Refactor small components with AI assistance
2. Build New Muscles
   • Learn to write clear, specific prompts
   • Practice validating AI output against architectural intent
   • Establish feedback loops to iteratively improve how you work with agents
3. Lead by Example
   • Share your journey and learning curves openly with your team
   • Create guidelines for AI-assisted development
   • Encourage shared libraries of effective prompts and system design patterns

Looking Ahead: The Next 18 Months

The rate of change in AI tooling and agentic architecture is exponential. Where you focus will depend on your context, but key foundations are universal:

• Emphasise architectural integrity over implementation correctness
• Evolve code review practices to include AI-generated contributions
• Maintain strong documentation of system boundaries and responsibilities
• Instrument your systems deeply—observability is the best early surface for experimenting with agents
• Make sure you’re thinking about data effectively

For more, see my articles that focus more deeply on:

1. Agentic AI
2. Data as a strategic asset

Conclusion

Agentic AI won’t replace seasoned engineering leadership. But it will reward those who retool their experience for this new collaborative frontier.

We’re not just getting better tools—we’re gaining new collaborators. Our role is shifting from crafting every line of code to directing and curating a human–AI partnership.

Start small. Stay curious. Share what you learn.

Let’s shape what’s next.

Back in 2017, I tweeted that

“Functional specialisation of individuals destroys team agility. Functional specialisation of teams destroys organisational agility.”

Years later, this observation has only grown more relevant as organisations struggle with digital transformation and rapid market changes.

Reflecting on the coming power of agentic AI driven workflows, I think the deep understanding of these concepts and how to implement them is going to be the primary differentiator between companies that surf the coming tide, and those that sink between the waves.

The Allure of Specialisation

Specialisation is seductive. It feels efficient. It’s easy to measure and manage. It maps neatly to job descriptions, reporting lines, and procurement contracts.

We’re drawn to specialisation because it promises efficiency. It’s comfortable. It’s measurable. But this comfort comes at a cost that’s often invisible until it’s too late.

The Hidden Costs

1. Knowledge Silos: When individuals become too specialized, knowledge doesn’t flow naturally through the team

2. Dependency Chains: Work gets stuck waiting for the “right person” to be available

3. Reduced Learning: Team members stop growing outside their specialty

4. Conway’s Law Effects: Our software begins to mirror our organisational divisions

Breaking Free

Let’s unpack this across three levels: individual development, team formation, and organisational design.

1. Individuals: Go Beyond T-Shaped

The T-shaped metaphor—deep in one area, broad in others—is better than hyper-specialisation, but still too static and shallow for today’s context. Instead:

• Think in terms of versatility over shape: What situations can this person adapt to? How quickly can they move across cognitive and technical domains?
• Encourage comb-shaped growth: multiple areas of depth with connected learning paths.
• Prioritize judgment, reasoning, and systems thinking over specific technical skills. Today’s full-stack engineer will be tomorrow’s full-context collaborator.

Key enablers:

• Internal mobility
• Role fluidity inside teams
• Learning budgets and time structures (e.g., pomodoro-for-learning commitments)

2. Teams: Optimize for Flow and Evolution

Team Topologies provides a much better mental model than functional roles. Instead of asking “What specialists do we need?” ask:

“What team structures optimize for flow of change, clarity of ownership, and fast feedback?”

Design teams as long-lived, cross-functional units with aligned purposes, shaped for interaction modes:

• Stream-aligned teams focused on delivering value end-to-end
• Enabling teams that lift capabilities of others
• Complicated subsystem teams owning complex internals (used sparingly)
• Platform teams that remove cognitive load by providing reusable abstractions

Avoid persistent handoffs between teams. Instead, embed, swarm, or pair across boundaries when specialisation is required.

Key enablers:

• Clearly defined interaction modes
• Lightweight team APIs (ways of working and communicating)
• Time-boxed co-working over long handovers

3. Organizations: Shape Context, Not Control

Organizations must stop trying to predefine the “perfect shape” of talent or teams and instead:

Create environments of aligned autonomy: where decisions happen close to the problem, but with strategic coherence.

This means:

• Incentives reward collaborative delivery and capability building, not heroics or individual depth
• Technical strategy is expressed through decision fitness functions and guardrails, not rigid roadmaps
• Visibility into why decisions are made (not just what is decided) is shared across levels to create organisational learning

Key enablers:

• Sociotechnical architecture reviews, not just code reviews
• Fitness functions that encode architectural and operational priorities
• Narrative transparency: shared understanding of priorities, trade-offs, and strategic goals

Moving Forward

The key is understanding that agility isn’t about doing things faster - it’s about maintaining options whilst still moving rapidly and iteratively in a desired direction. Functional specialisation reduces options, which directly impacts our ability to adapt and respond to change.

Remember: You can’t predict a complex work system using metrics, but you can use metrics to learn how to change a system to be more predictable.

In a previous post, I explored how functional specialisation — at various levels of an organisation — creates hidden organisational debt that slows it down and reduces optionality. Slow delivery, brittle systems, and increasing misalignment between how we structure our people and how we want our systems to behave.

But that analysis assumed a world of mostly human actors.

Now, with the accelerating rise of agentic AI—tools and systems capable of taking initiative, decomposing goals, and coordinating their own workflows—we’re entering a new design space.

And we’re going to need new organisational metaphors, constraints, and incentives to match.

This is still an emerging field. But the early signals are clear: as agentic tooling becomes more capable, the way we structure our organisations will be a major determining factor on our ability to amplify (and constrain) their potential.

Those companies who fail to adapt will get rapidly replaced by those that do, with new kinds of companies emerging that were not possible before.

What thinking that we currently hold tightly to will still be valid, and what will need to let go of or change?

Why Agentic AI Changes the Game

Most automation in the last few decades has operated in a task-execution paradigm: software did narrow things well, often behind the scenes, often embedded in business processes that humans designed and oversaw.

Agentic AI changes that. We now have systems that:

• Take broad goals as input and break them into subtasks
• Select tools, APIs, and data sources
• Manage state across long-lived flows
• Coordinate with humans (or other agents) as collaborators, not just passive endpoints

That changes the role of the human, and shifts the boundary between “doing the work” and “designing the system that does the work.”

It also raises urgent questions:

• Who is accountable for the outcomes of an AI-augmented workflow?
• How do we ensure coordination across agents and humans when both are initiating change?
• What does “team” even mean when a portion of it isn’t human?

The End of the Knowledge Monopoly

Traditional organisational hierarchies were built on the premise that knowledge and expertise flow from the top down. Functional specialists held monopolies over their domains. But agentic AI is about to shatter this model:

1. Expertise Becomes Ambient: When AI can provide expert-level insights in real-time, the value shifts from holding knowledge to knowing how to apply it effectively

2. Horizontal Knowledge Flow: Information and learning will flow sideways through AI-mediated channels, bypassing traditional vertical structures

3. Role Fluidity: Fixed functional roles become less relevant when AI can rapidly scaffold domain knowledge

New Organisational Primitives

Instead of organising around functional specialties, companies may need to organise around:

1. Context Pods: Small, cross-functional groups with deep shared context about specific business domains

2. Learning Loops: Structures optimised for rapid experimentation and feedback between humans and AI

3. Intent Networks: Loose configurations of humans and AI agents aligned around clear outcomes rather than prescribed processes

The Human Advantage

The key differentiator becomes our uniquely human capabilities:

1. Intent Setting: Defining meaningful direction and purpose

2. Context Synthesis: Combining multiple viewpoints into coherent understanding

3. Social Orchestration: Building trust and alignment across human-AI teams

The organisations that thrive won’t be those with the best AI, but those who best understand how to create environments where humans and AI amplify each other’s strengths.

What Principles Still Hold?

Despite the hype and change, some organisational design truths will most likely remain remarkably resilient.

1. Sociotechnical alignment still matters

The core principle from Conway’s Law still applies: your systems will reflect your communication structures. If you introduce autonomous agents into brittle, siloed org structures, you’ll get fragmented, brittle AI usage.

2. Flow beats function

Designing for flow of change, feedback, and value will still outperform static hierarchies and handoff-heavy workflows. Agentic tools can enhance flow—but only if the environment supports it.

3. Clear purpose and context are essential

Agents need intent, scope, and constraints to act meaningfully. Humans do too. Making decision-making context legible at all levels—through goals, metrics, affordances, and APIs—remains essential.

What Might Need to Change?

This is where things get exciting—and uncertain.

1. From roles to capabilities to outcome-defined functions Traditional org charts based on stable roles are already under pressure. With agents in the mix, we’ll need to model work in terms of capability graphs, where both humans and AI components can be dynamically assigned to outcome-defined goals.

1. From roles to capabilities to outcome-defined functions

Traditional org charts based on stable roles are already under pressure. With agents in the mix, we’ll need to model work in terms of capability graphs, where both humans and AI components can be dynamically assigned to outcome-aligned goals.

Think: “Who can satisfy this intent under these constraints?” not “Whose job is this?”

2. Teams as dynamic coalitions

The idea of a “team” as a static group of people may give way to fluid coalitions of humans and AI agents assembled around a problem or opportunity. This could drive a move toward short-lived, goal-oriented micro-teams, orchestrated by platform or protocol.

3. Organisational structure becomes more recursive

As agents become internal developers, decision-makers, and testers, we may see orgs that are nested systems of delegation—humans designing high-level strategies and constraints, agents handling tactical execution within those boundaries.

This mirrors how effective engineering teams work today: strategy at the top, execution at the edge, with feedback loops across the system. The difference is who (or what) is at the edge.

4. Governance, ethics, and observability become foundational

When agents are taking actions with consequences, we need radically better observability, traceability, and accountability. These aren’t bolt-ons. They become the scaffolding of trustworthy organisational systems.

Final Thoughts: It’s Not Just About Tech, or making the right tool decisions

This shift isn’t just technical—it’s deeply human and organisational.

Just like DevOps and agile ways of working required rethinking culture, incentives, and collaboration, agentic AI will force a similar reckoning. It’s not enough to bolt agents into old workflows. We’ll need to rethink what work is, who does it, and how systems evolve over time.

The organisations that thrive in this next wave will be those that design for adaptability, not just efficiency.

That means:

• Investing in platforms that support dynamic assembly
• Building incentives around outcomes and learning, not control
• Focusing on resilience, feedback loops, and optionality

We’re just getting started. But the future of work won’t be shaped by AI alone—it’ll be shaped by how we choose to organise around it.

Building on previous posts about functional specialisation and agentic AI, let’s explore how the way we organise around data will become a primary differentiator between companies that thrive and those that struggle in the age of ambient intelligence.

The Traditional Approach Is Failing

Most organisations still treat data as primarily a technical challenge:

• They hire data engineers and architects
• They buy data platforms and tools
• They create data strategies focused on storage, processing, and analytics

But they’re missing the fundamental point: Data is a sociotechnical artifact. It’s continually created, modified, and consumed through human interactions, decisions, and organisational processes.

And in a world where increasingly intelligent agents are acting on our behalf, this data becomes the substrate of all action, judgment, and coordination. It is not just an asset. It’s the fuel, the language, and the terrain.

And yet, most organisations treat it like exhaust.

The Hidden Costs of Poor Data Design

Just as with functional specialisation, treating data as purely technical creates hidden organisational debt:

1. Knowledge Silos: Data becomes trapped in functional teams
2. Quality Decay: No clear ownership of data quality across its lifecycle
3. Lost Context: Critical business context gets stripped away
4. Conway’s Law Effects: Data structures mirror org charts, not business reality

Why This Matters Even More Now

We’re entering a phase where tools will no longer ask, “What do you want me to do?”
Instead, they will ask: “What outcome do you want? I’ll figure out the steps.”

This shift—from tool to agent—makes the quality, structure, and accessibility of data a first-order concern.

If your data is fragmented, siloed, poorly contextualised, or hidden behind org boundaries and tribal knowledge, then:

• Your agents will hallucinate, misfire, or underperform
• Your teams will drown in translation layers and rework
• Your strategy will be constrained by your least tractable data bottlenecks

Shifting the Paradigm

Instead of asking “How do we manage our data?”, we need to ask:

“How do we design our organisation to treat data as a first-class strategic asset?”

This reframing leads to different approaches:

1. Data as a Product

• Treat datasets like products with clear owners and stakeholders
• Apply product management principles to data lifecycle
• Focus on user needs and experiences, not just technical specifications

2. All Teams are Data Teams

• Embed data expertise across the organisation
• Consider creating “data pods” in that combine business and technical skills in key areas
• Focus on outcomes over output

3. Data Governance as Enabling Constraint

• Design governance that enables innovation while maintaining quality
• Create clear decision rights and accountability
• Build feedback loops between data producers and consumers

Organisational Design Implications

The real problem is organisational, not just technical. This shift requires rethinking several aspects of how we organise.

We’ve spent decades building pipelines, lakes, warehouses, lakehouses, and catalogs. And yet most organisations still struggle to answer basic cross-cutting questions about their operations, customers, and systems.

Why?

Because this isn’t a tooling problem. It’s a sociotechnical and organisational design problem.

You can’t layer a “data strategy” on top of an organisation that wasn’t designed to create, share, and learn from data as part of its core functioning.

Let’s point out the common anti-patterns.

1. Ownership without responsibility

Data is generated everywhere but owned nowhere. Product teams create it, infra teams store it, analysts clean it, and no one feels responsible for its lifecycle, meaning, or reusability.

2. Data work is treated as cost, not capability

Too often, work on data quality, modeling, documentation, and lineage is seen as overhead—something you squeeze in “if there’s time.” Meanwhile, delivery pressure drives shortcuts that poison long-term optionality.

3. Incentives reinforce fragmentation

Different departments optimise for their own goals. Sales logs data one way, support another. Product managers hoard insights. Legal locks down access. Data becomes a reflection of org silos, not a unifying strategic fabric.

4. Tooling abstracts away the real work

Modern data stacks can give a false sense of progress. You can have perfect pipelines flowing into a state-of-the-art lakehouse—and still not know how many active users you have last quarter, or why a metric changed.

Treating Data as a Strategic Asset Means…

We need to go far beyond governance frameworks and tech stacks. We need to design our organisations as data-fluent systems.

That means:

1. Team-level accountability for data quality and meaning - Teams that generate data must also be responsible for its semantic clarity, quality over time, and downstream utility. Think data as a product, but culturally embedded—not a side project.

2. Cross-functional alignment around information needs - Instead of each team defining its own schema of reality, we need shared models of the domain—kept alive through active stewardship, not doc rot. Context pods or capability squads should include data stewards by design.

3. Organising for data discoverability and intent alignment - Good metadata and cataloging are helpful—but only when paired with social and technical systems that make data legible to both humans and agents. This means:

• Discoverability by purpose, not just source
• Documentation embedded in usage contexts
• Feedback loops between consumers and producers

4. Building data fluency into every role

We need to stop outsourcing “data work” to a separate class of specialists. Instead, every knowledge worker should have a baseline of data fluency, and every team should have the means to reason about their data and its use.

5. Team Structures

Move from centralised data teams to:

• Embedded data capabilities in teams
• Data platform teams that focus on provide enabling infrastructure
• Data governance networks that span the organisation and focus on securing the flow of data, not blocking it

6. Incentive Alignment

Create incentives that reward:

• Data quality over quantity
• Reuse over reinvention
• Context preservation over pure efficiency

Agentic AI Raises the Stakes

Once you introduce agentic systems into your organisation, every flaw in your data becomes a multiplying factor in systemic fragility.

• Agents need reliable state to plan and act
• They need well-modeled domains to reason about
• They need feedback signals to learn

This isn’t just about building the right APIs. It’s about designing your organisation so that high-quality, context-rich data emerges as a natural byproduct of how work gets done.

The Path Forward

Success requires focusing on three key areas:

1. Organisational Clarity

• Clear ownership and decision rights
• Explicit data quality standards
• Visible value chains from data to decisions

2. Technical Architecture

• Platforms that enable rather than constrain
• Tools that preserve context
• Infrastructure that promotes discovery and reuse

3. Cultural Transformation

• Treating data as a shared asset
• Building data literacy at all levels
• Creating psychological safety around data quality issues

Practical Steps

1. Map Your Data Ecosystem
   • Identify key data products
   • Map data flows and dependencies
   • Understand decision points and value creation
2. Design for Evolution
   • Create clear interfaces between teams
   • Build feedback loops into processes
   • Enable rapid experimentation and learning
3. Invest in Capabilities
   • Technical platforms that enable collaboration
   • Training and development programs
   • Metrics that matter

Final Thoughts

As data becomes more central to how organisations operate and compete, the ability to design effective sociotechnical systems around data will become a critical capability.

Managing data as a strategic asset is not just a CIO or CDO problem. It’s a whole-organisation design challenge.

The organisations that thrive won’t be those with the best technology, but those who best understand how to create environments where humans and machines can effectively collaborate around data to create value.

Focus on

• Rewriting incentives to reward data reusability, not just delivery
• Building sociotechnical systems that make high-leverage data work visible and valuable
• Embedding data thinking into org design, team structures, and operational rhythms

Remember: You can’t solve sociotechnical problems with purely technical solutions. The way forward is through intentional design of the entire system—people, process, and technology.

In the age of agentic AI, your data is no longer just an asset—it’s the terrain your organisation navigates, the medium your AI thinks in, and the context your humans coordinate through.

Organisations that learn to flow with their data will outlearn, out-adapt, and outlast those that don’t.

To the modern eye, Zoos are a depressing place. They are a collection of cages and enclosures housing animals that are not free to roam as they would be in the wild. They’re a reflection of a society that valued entertainment above the welfare of the animals. Johannesburg Zoo is no different, and perhaps more so than most because tourists arrive here from around the world to see our beautiful wildlife in it’s natural habitat. Yet the zoo premises are old, the maintenance has fallen behind and the priority of the city municipality is, understandably elsewhere.

And so while Johannesburg needs more world class tourist attractions to generate revenue for the city to invest in upliftment projects, the zoo situated on prime property within the city exists in a state of increasing disintegration and decay. If nothing changes, the zoo will continue to decline and become a sad symbol of a city that once aspired to greatness.

However, if the zoo’s future could be aligned with the real and challenging priorities of the people and municipality of Johannesburg, the right strategy could create something we, as residents of Johannesburg, take immense pride and satisfaction in and people come from around the world to visit.

I put this business plan together to explore how this could be achieved, and then published it because it might inspire others to think about what could be possible…

Business Plan for the Future of the Johannesburg Zoo

Date: October 2024 —

Executive Summary

The Johannesburg Zoo, owned by the City of Johannesburg through Johannesburg City Parks and Zoo (JCPZ), is poised for a transformative journey. This business plan outlines a strategy to reinvent the zoo as a dynamic, self-sustaining institution that not only serves as a hub for practical education and community engagement, it also provides a platform for the conservation of wildlife whilst generating income for the youth of the city.

By implementing a structured, three-tiered internship program and pivoting towards conservation efforts, the result will enhance its value to the city, promote positive social impact, and ensure long-term sustainability.

This plan will transform the zoo into a world class destination that will be a key tourist attraction for locals and international visitors alike.

Key Value Proposition to Current Owners

Enhanced Public Image and Reputation

• Leadership in Conservation: Positioning the zoo as a leading wildlife rehabilitation center enhances the city’s reputation in environmental stewardship.
• Youth Empowerment: Offering valuable internship opportunities showcases the city’s commitment to addressing youth unemployment and education.
• Community Engagement: Hosting cultural events and educational programs strengthens community ties and promotes social cohesion.

Financial Sustainability

• Self-Sustaining Model: Diversified revenue streams ensure the zoo operates without requiring additional funding from the city budget.
• Economic Development: Increased visitor numbers and events stimulate local economy and create indirect job opportunities.

Alignment with City Priorities

• Education and Job Creation: Directly addresses key governmental objectives, enhancing the city’s social development goals.
• Environmental Conservation: Supports biodiversity and aligns with national and international conservation commitments.

Company Description

• Name: Johannesburg Zoo
• Ownership: City of Johannesburg through Johannesburg City Parks and Zoo (JCPZ)
• Mission: To become a leading institution in wildlife rehabilitation while empowering youth through a comprehensive, hands-on internship program.
• Vision: To foster a sustainable future where conservation, education, and community enrichment thrive in harmony.

Market Analysis

Economic Climate

• Youth Unemployment: High rates necessitate innovative solutions for job creation and skill development.
• Educational Demand: Graduates seek practical experiences to complement their academic qualifications.

Political Climate

• Governmental Support: Initiatives that promote education, employment, and conservation are likely to receive backing.
• Regulatory Environment: Favorable policies for environmental conservation and youth development programs.

Social Climate

• Community Needs: Strong desire for positive social initiatives and accessible cultural activities.
• Public Interest: Growing engagement in environmental issues and support for sustainable practices.

Competitive Landscape

• Unique Offering: No other local institution combines wildlife rehabilitation with a structured, multi-year internship program that includes cultural event management.
• Johannesburg as a tourist destination The city has a poor reputation as a tourist destination, but the zoo could be the jewel in the crown.

Organization and Management

Core Professional Management Body

• Composition: Specialists in zoology, education, marketing, finance, event management, and operations.
• Roles and Responsibilities:
  • Oversee daily operations and strategic direction, primarily through the internship program.
  • Mentor interns across all program levels.
  • Facilitate partnerships with academic institutions and stakeholders.
  • Ensure compliance with regulatory requirements and industry best practices.

Internship Program Structure

First Year: Foundational Experience

• Eligibility: Recent matriculants or university graduates (within the last two years).
• Roles:
  • Field-specific positions (e.g., marketing, finance, animal care).
  • Non-optional rotational duties in animal and enclosure care and garden maintenance to promote a holistic understanding & appreciation of hard work and responsibility.
• Development Focus: Practical skills application, teamwork, basic professional responsibilities, pride in a job well done.

Second Year: Management Development

• Selection Criteria: Performance, attendance, leadership potential from the first year.
• Roles:
  • Management interns overseeing first-year participants.
  • Focused on team leadership and operational management.
• Development Focus: Leadership skills, decision-making, supervisory experience.

Third Year: Strategic Leadership

• Selection Criteria: Outstanding performance and strategic thinking in the second year.
• Roles:
  • Strategic interns collaborating with the core management team
  • Involvement in high-level planning and strategic initiatives to evolve and grow the zoo into a world class destination.
• Development Focus: Strategic management, organizational leadership, project planning.

Services and Programs

Wildlife Rehabilitation and Conservation

• Core Focus: Transition from traditional animal exhibits to active wildlife rehabilitation and breeding programs for endangered species.
• Objectives:
  • Rehabilitate local wildlife for reintroduction into natural habitats.
  • Breed endangered species to support conservation efforts.
• Benefits to Owners:
  • Enhances the city’s environmental conservation profile.
  • Opportunities for partnerships with conservation organizations.

Educational and Internship Programs

• Three-Tiered Structure: Provides progressive learning and leadership opportunities.
• Collaborations: Partnerships with universities and colleges for research and practical training.
• Benefits to Owners:
  • Addresses unemployment through skill development.
  • Positions the city as a leader in innovative education models.

Cultural Events and Public Engagement

• Artistic Endeavors: Support for interns from creative disciplines to create installations and performances on-site.
• Community Events: Hosting musical, dramatic, and cultural events to attract visitors.
• Benefits to Owners:
  • Enhances community relations and cultural richness.
  • Generates additional revenue and increases public interest in the zoo.

Marketing and Sales Strategy

Target Audiences

• Internship Applicants: Attracting high-caliber candidates for the internship programs.
• Academic Institutions: Establishing the zoo as a preferred partner for research and student development.
• General Public: Encouraging zoo visits and event attendance.
• Donors and Sponsors: Securing funding and support from aligned organizations.

Marketing Initiatives

• Intern-Driven Campaigns: Empowering marketing interns to develop and execute promotional strategies.
• Digital Outreach: Enhancing online presence through social media and an interactive website.
• Community Programs: Engaging local schools and organizations through workshops and outreach activities.
• Global Partnerships: Attracting international interest and collaboration for funding and knowledge exchange.

Operational Plan

Facility Management

• Upgrades and Renovations: Modifying existing structures to support rehabilitation efforts and educational spaces.
• Maintenance Protocols: Implementing regular maintenance schedules managed by interns under professional supervision.

Animal Care and Welfare

• Holistic Involvement: All interns participate in animal care, fostering a culture of respect and responsibility.
• Expert Oversight: Experienced veterinarians and zoologists lead animal welfare programs.

Event Planning and Execution

• Interdepartmental Collaboration: Interns from various disciplines work together to plan and host events.
• Professional Development: Real-world experience in event management and operations.

Financial Plan

Revenue Streams

• Admissions and Event Tickets: Income from zoo visits and hosted events.
• Donations and Grants: Financial support from individuals, corporations, and NGOs.
• Merchandise and Concessions: Sales of branded goods and art created by interns.
• Sponsorships and Partnerships: Funding through corporate sponsorships and collaborative projects.

Cost Structure

• Intern Stipends: Competitive compensation to attract top talent and support diversity.
• Operational Expenses: Costs associated with animal care, facilities, utilities, and program implementation.
• Reinvestment Strategy: All profits are reinvested into the zoo’s programs, enhancing sustainability and impact.

Financial Projections

• Break-Even Point: Anticipated within the first two years due to diversified revenue and cost management.
• Growth Forecast: Steady increase in revenue aligned with program maturity and brand recognition.

Implementation Timeline

The first milestone will be to prove the concept of the internship program and the value it brings to the youth of the city, whilst maintaining the current way the zoo operatates and is maintained. Once the concept is proven, as the second and third year interns become available, the focus will shift to scaling the program and evolving the zoo into a world class destination.

Phase 1: Foundation Building (Months 1-3)

• Establish Management Team: Recruit and onboard core professionals.
• Develop Programs: Finalize internship curricula and operational protocols.
• Initiate Partnerships: Secure collaborations with educational institutions and potential sponsors.

Phase 2: Seed Program Launch (Months 4-6)

• Marketing Rollout: Begin promoting the internship program and upcoming events.
• Facility Preparations: Complete necessary renovations and setup for new programs.
• Select Interns: Conduct a rigorous selection process for the first cohort of interns.

Phase 3: Operational Activation (Months 7-12)

• Internship Kickoff: Start the first year of the internship program.
• Program Execution: Implement wildlife rehabilitation projects and commence event planning.
• Community Engagement: Host initial public events and educational workshops.

Phase 4: Growth and Evaluation (Year 2 and Beyond)

• Program Advancement: Progress interns through second and third-year roles.
• Continuous Improvement: Regularly assess programs for effectiveness and make necessary adjustments.
• Expansion Opportunities: Explore additional services, partnerships, and revenue channels.

Community and Social Impact

Educational Advancement

• Skill Development: Equip interns with practical experience, enhancing their employability.
• Leadership Training: Prepare future leaders with real-world management and strategic planning skills.

Conservation Contributions

• Biodiversity Support: Active participation in the rehabilitation and breeding of local wildlife.
• Environmental Education: Raise public awareness about conservation issues.

Cultural Enrichment

• Accessible Events: Provide affordable cultural experiences for the community.
• Artistic Platform: Offer a venue for emerging artists to showcase their talents.

Conclusion

This business plan offers a comprehensive strategy that significantly enhances the value of the Johannesburg Zoo to its current owners, the City of Johannesburg and JCPZ. By transforming the zoo into a self-sustaining institution focused on conservation, education, and community engagement, it aligns with the city’s objectives and provides measurable benefits in social impact, financial sustainability, and public reputation. The proposed initiatives ensure that the zoo not only thrives as a landmark institution but also contributes meaningfully to the city’s development and the well-being of its residents.

Appendices

Appendix A: Risk Analysis

Financial Risks

• Revenue Variability: Fluctuations in visitor numbers or donations could impact financial stability.
  • Mitigation Strategies: Diversify income sources, build reserve funds, and implement flexible budgeting.

Operational Risks

• Program Scalability: Managing a large number of interns may present logistical challenges.
  • Mitigation Strategies: Establish clear management hierarchies, provide thorough training, and maintain optimal intern-to-supervisor ratios.

Reputational Risks

• Public Perception: Resistance to changes in zoo operations or negative publicity from any incidents.
  • Mitigation Strategies: Maintain transparent communication, engage in proactive community outreach, and uphold high standards of animal welfare and safety.

External Risks

• Economic Downturns: Economic challenges could affect attendance and funding.
  • Mitigation Strategies: Adjust pricing strategies, enhance online and virtual engagement options, and seek alternative funding sources.

Appendix B: Organizational Chart

• Visual representation of management structure, including the core professional team and the internship tiers.

Appendix C: Financial Projections

• Detailed financial statements, including projected income, expenses, and cash flow over the next five years.

Appendix D: Market Research Data

• Statistics and analyses supporting the market analysis section, including data on youth unemployment, educational trends, and public interest in conservation.

Appendix E: Letters of Support

• Endorsements from educational institutions, government agencies, conservation organizations, and community leaders.

Prepared by: [Your Name], [Your Position] Contact Information: [Phone Number], [Email Address]

Tim hosts the DevJourney podcast, with the main theme of having an open ended conversation with someone who’s made a career from being a software developer.

Blurb from Tim on our convo…

Kevin started his story in the 80 & the 90s but quickly said, “Coding found me more than I found it.” We then discussed his (very early) Bootcamp and how he went from one job to the next, slowly feeling less incompetent. We talked about ADHD, networking, people & organizational problems, and environment variables. We then dug our heels into eXtreme Programming and the Spine model: Needs, Values, Principles, Practices, and Tools, and how it helps us create better teams and organizations.

The episodes of the podcast include some fascinating people with very interesting stories, I felt it was a privileged to be added to the list.

You cannot talk to a team. You cannot ask a team a question. Teams have no feelings or goals.

But but but…favourite soccer team, my team!

Sorry. They don’t exist.

Where a team can be observed, what you are actually seeing is:

• Each individual’s behaviours,
• in emergent response to what is going on around them,
• as processed through their archetype of what a team should be
• with their latest individual mental model of what their team currently is, and
• their individual role and purpose relative to the team at the given moment.

Protip: Stop talking to your teams. Start talking to your people.