The Future Operating Model for IBM Garage in the Age of Agentic AI
From Agile at Scale to Agentic Value Orchestration
For years, IBM Garage has helped organizations move from idea to outcome through a disciplined model of co-creation, co-execution, and co-operation. It gave enterprises a practical way to connect strategy, design, engineering, business value, governance, and adoption into one operating model.
But the world around that model is changing.
Agentic AI is not simply another productivity tool. It is changing how work is decomposed, specified, built, tested, governed, supported, and improved. It is changing the economics of digital delivery. It is changing the shape of squads. And it is changing the role of humans from people who manually perform every activity to people who define intent, set direction, provide judgment, manage risk, ensure quality, and orchestrate value.
The future IBM Garage operating model should not abandon what made Garage successful. In fact, most of the operating model should remain. The portfolio structure, governance boards, investment controls, chapters, communities of practice, design authority, value orchestration, and adoption mechanisms become even more important.
What changes most is the squad.
The future Garage will be less about large delivery teams executing tasks manually, and more about small, sticky, highly capable human squads directing AI agents through well-defined goals, specifications, standards, and guardrails.
The operating model evolves from:
Human-heavy agile delivery
to:
Human-led, AI-accelerated, value-orchestrated delivery
And that distinction matters.
What Should Not Change
A common mistake in thinking about agentic organizations is assuming that AI will replace the operating model. It will not.
AI can accelerate activities, generate insights, draft artifacts, write code, test solutions, produce options, monitor signals, and support decision-making. But it should not replace the governance, accountability, prioritization, value discipline, and human judgment required to run an enterprise technology organization.
The future IBM Garage operating model still needs the same core interface structures.
The Investment Board remains essential because funding choices do not disappear in an AI-enabled organization. In fact, they become more important. If AI makes it easier to generate ideas, prototypes, and solution options, the enterprise needs an even stronger mechanism to decide where capital should go, which business opportunities deserve investment, and which ideas should stop.
The Garage Board also remains essential. It continues to manage impediments, prioritization decisions, capability constraints, capacity trade-offs, location strategy, delivery risks, and cross-portfolio escalation. AI can surface issues faster, but leadership still needs to decide what to do about them.
The Core Garage Team remains critical. The Garage Leader, Value Orchestration Lead, and Delivery Excellence Lead continue to provide the connective tissue across business, technology, delivery, governance, and measurable outcomes.
The SME Bar, chapters, and communities of practice also remain. Engineering, design, product, industry, data and AI, security, cloud, QA, Agile, DevOps, and architecture chapters may become AI-augmented, but they should not disappear. These groups set standards, define patterns, review exceptions, assess risk, and ensure consistency across squads and products.
This is especially important in an agentic organization. When AI agents are generating designs, code, tests, documentation, and operational recommendations, standards matter more, not less.
The future operating model needs more discipline around reusable specifications, engineering guidelines, architectural patterns, design systems, testing policies, security controls, and value metrics. AI agents can help enforce those standards, but humans must remain accountable for accepting risk, approving exceptions, and evolving the standards over time.
What Changes Most: The Squad
The current Garage operating model is built around different squad archetypes across the product lifecycle: Co-create squads, Co-execute squads, and Co-operate squads. These squads are aligned to portfolios and supported by portfolio management, agile coaching, design authority, chapters, and governance ceremonies.
That structure still makes sense.
But the composition and operating rhythm of the squads will change significantly.
In the future, the standard squad will be much smaller. The model I believe will dominate is a three-person AI-augmented squad:
- Product Owner
- Forward Deployment Engineer
- Designer / Design Researcher
This does not mean every squad will always have exactly three people. Complex systems may require multiple pods working in parallel. Highly regulated environments may require additional risk, security, compliance, or change specialists. But the default unit of execution becomes smaller, more senior, more cross-functional, and much more AI-enabled.
The Product Owner defines business intent, prioritizes outcomes, validates value, and keeps the squad anchored in business impact.
The Forward Deployment Engineer combines architecture, full-stack engineering, solution shaping, integration thinking, and AI-agent direction. This role is not simply a developer. It is a builder, architect, translator, and orchestrator of agentic execution.
The Designer or Design Researcher ensures that the work remains human-centered. This role shapes the target experience, validates user needs, protects usability, and ensures that adoption is designed into the solution rather than treated as a post-launch activity.
The major shift is that the squad is no longer sized around the assumption that every task must be manually performed by humans. Instead, the squad is sized around the assumption that humans define intent, direction, quality, context, and decisions while AI agents accelerate analysis, generation, testing, documentation, and delivery.
Co-create: Accelerated, But Still Human-Led
The Co-create phase should change the least in terms of human involvement, but it should change significantly in speed.
Co-create is where strategy becomes a value framework. It is where teams identify observations, formulate problem hypotheses, validate pain points, map personas and journeys, generate solution hypotheses, and shape business opportunities.
AI can accelerate almost every one of these steps.
An interview transcript can be converted into observations. Observations can be clustered into themes. Themes can become problem hypotheses. Problem hypotheses can be tested against additional research. Pain points can be drafted. Personas and journey maps can be generated. Solution ideas can be proposed. A business opportunity canvas can be pre-populated.
Platforms like IBM Garage Experience can already support this kind of AI acceleration.
But Co-create should not become fully autonomous.
The reason is not that AI cannot generate useful ideas. It can. The reason is adoption.
The envisioning moment still matters. People need to come together, review the challenges, understand the user journeys, debate the solution options, and make decisions collectively. That shared decision-making creates ownership. Ownership creates adoption.
If AI generates the solution and the organization is simply told to implement it, adoption will suffer. People will not know why decisions were made. They will not see themselves in the solution. They will feel that change was done to them, not with them.
So the future Co-create squad should be smaller and faster, but still deeply human.
AI should prepare the room.
Humans should make the meaning.
Co-execute: The Biggest Transformation
Co-execute is where the operating model changes most dramatically.
In the traditional model, Co-execute requires larger squads because design, development, testing, architecture, quality, DevSecOps, and release activities are performed largely by humans. In an agentic model, much of this work becomes specification-driven and AI-assisted.
The squad’s job shifts from manually creating every deliverable to defining the right specifications, rules, guardrails, acceptance criteria, user experience requirements, architecture constraints, and engineering standards.
The Forward Deployment Engineer becomes central.
This role works with the Product Owner to understand what needs to be built, works with the Designer to ensure the target experience is clear, and then directs AI agents to generate working deliverables. From the first iteration, the squad should have something tangible to inspect, test, refine, and improve.
In this model, execution becomes much more parallel.
A complex product can be decomposed into components, services, experiences, workflows, data products, APIs, or subsystems. Multiple small pods can work in parallel, each directing AI agents against a clear specification. The challenge is no longer only “how many developers do we need?” The challenge becomes “how do we coordinate parallel agentic delivery without losing architectural coherence, experience consistency, quality, security, or value alignment?”
That is why governance does not go away.
Product increment planning still matters. Architecture review still matters. Design authority still matters. Portfolio synchronization still matters. Standards still matter. Quality still matters.
AI changes the speed of execution. It does not remove the need for orchestration.
Co-operate: From Manual Support to Agentic Operations
Co-operate will also be transformed.
Once a product has achieved product-market fit or enterprise adoption fit, the question becomes how to scale, harden, operate, maintain, and continuously improve it.
In the current model, Co-operate often requires meaningful development, support, reliability, quality, and maintenance capacity. In the future, much of this work will be AI-assisted or agent-driven.
AI agents can monitor production signals, identify incidents, summarize root causes, propose fixes, generate regression tests, update documentation, detect usage patterns, recommend improvements, and support service reliability teams.
But again, humans remain accountable.
The Co-operate squad becomes a small human team that governs agentic operations. It ensures that agents are working within approved guardrails, that incidents are handled appropriately, that changes are validated, that user experience does not degrade, and that business value continues to be realized.
The operating model shifts from “humans perform support activities” to “humans supervise, improve, and govern intelligent operational systems.”
This is where the Value Office becomes especially important.
The point of AI-enabled operations is not just to reduce support cost. It is to protect and grow the value of the product. That means tracking adoption, usage, reliability, benefit realization, productivity improvement, cost reduction, customer experience, risk reduction, and financial outcomes.
Co-operate is not the end of Garage.
It is where the value curve must prove itself.
The Portfolio Model Still Holds
The future Garage operating model should remain portfolio-based.
Portfolios create business alignment. They allow squads to build domain knowledge. They create continuity. They help organize investment, capability, capacity, governance, prioritization, and value tracking around meaningful business areas.
What changes is how squads are managed inside the portfolio.
The future model should use sticky pods.
A small squad should not be assembled and disassembled for every initiative. It should develop domain knowledge, product intuition, stakeholder relationships, and technical familiarity over time. It should move from opportunity to opportunity as a pod, not as a set of disconnected resources.
This matters even more in an AI-enabled organization.
The more AI accelerates delivery, the more context becomes the differentiator. A sticky squad that understands the business domain, enterprise architecture, user needs, data landscape, operational constraints, and value drivers will outperform a temporary group of individuals, even if both have access to the same AI tools.
Agentic delivery rewards context.
Portfolios are how that context is preserved.
The Value Office Becomes More Important
In an AI-enabled enterprise, there will be no shortage of ideas.
There will be too many.
AI will make it easier to generate use cases, prototypes, concepts, mockups, process redesigns, and automation opportunities. The risk is that organizations will confuse activity with value.
That is why the Value Office becomes one of the most important parts of the future Garage operating model.
The Value Office must ensure that every initiative has a clear value hypothesis, measurable outcomes, baseline metrics, target metrics, ownership, and a benefit realization path. It must ensure that the organization is not investing in attractive ideas that do not produce meaningful business impact.
This is especially important because many AI initiatives fail to reach production or fail to produce measurable value after launch. The future Garage model needs strong value orchestration from the start:
- What business outcome are we targeting?
- What value lever are we activating?
- What metric will prove progress?
- What is the baseline?
- What is the expected target?
- When should value start to appear?
- Who owns realization after launch?
- What happens if the initiative is off track?
AI can help monitor these signals, but humans need to govern the value conversation.
The Value Office becomes the mechanism that keeps the agentic organization honest.
The Design Authority Becomes More Important, Not Less
In many organizations, governance is viewed as a bottleneck. In an agentic organization, governance must become an accelerator.
The Design Authority should evolve into a more active, AI-enabled quality and standards system.
It should include representatives from engineering, product, design, data and AI, security, cloud, QA, architecture, industry, Agile, DevOps, and relevant technology Centers of Competency.
Its role is no longer just to review finished designs or approve exceptions. It must define the standards that AI agents use to do the work.
That includes:
- Architecture patterns
- Engineering standards
- Security controls
- Design systems
- UX principles
- Data standards
- API guidelines
- Testing policies
- Documentation requirements
- Responsible AI guardrails
- Reusable prompts and specifications
- Quality gates for agent-generated deliverables
The Design Authority may use AI agents to perform first-level reviews. For example, an architecture review agent could check a solution against enterprise patterns before the human board meets. A security agent could flag policy exceptions. A design agent could detect inconsistency with the design system. A QA agent could assess test coverage.
The human Design Authority then focuses on exceptions, trade-offs, risks, and decisions.
Get Dr. Mohamed El-Refai’s stories in your inbox
Join Medium for free to get updates from this writer.
This is how governance becomes faster without becoming weaker.
Ceremonies Still Matter
Agentic AI does not eliminate ceremonies. It changes what ceremonies are for.
Product Increment Planning still matters because humans must decide which features, outcomes, and usable deliverables should be grouped into each iteration. AI can break features into stories, tasks, tests, and execution plans, but humans need to shape the product increment around value and adoption.
Portfolio Sync still matters because multiple business opportunities will move through the funnel at different speeds. Portfolio leaders need to synchronize priorities, dependencies, capacity, capability, and investment decisions.
Capital, Capability, and Capacity Planning still matters because humans remain part of the system. Even if AI agents perform more work, organizations still need scarce human expertise to define intent, manage risk, approve exceptions, shape architecture, and engage stakeholders.
Impediment Management becomes even more important because blockers in an agentic model may include data access, environment access, model restrictions, security approvals, integration constraints, unclear specifications, or unresolved business decisions.
Enterprise Architecture Review still matters because AI-generated solutions must align with the enterprise architecture, technology strategy, data standards, and security posture. AI can accelerate the review process, but accountability remains human.
Investment Board governance still matters because the enterprise must continuously decide whether to continue, pivot, scale, or stop initiatives based on value evidence.
RAID management — risks, assumptions, issues, and dependencies — still matters because AI does not remove uncertainty. It may actually increase the need for transparency because work is moving faster.
The ceremonies remain, but the preparation, analysis, documentation, and review workflows become AI-assisted.
The To-Be Operating Model
The future IBM Garage operating model can be described as a portfolio-based, value-orchestrated, agentic delivery system.
At the top, the Investment Board controls funding, value expectations, and strategic allocation.
The Garage Board manages prioritization, impediments, capacity, capability, delivery health, and cross-portfolio decisions.
The Core Garage Team provides leadership across Garage operations, value orchestration, delivery excellence, and adoption.
The Value Office defines value hypotheses, metrics, baselines, targets, realization plans, and benefit tracking.
The Design Authority defines the standards, patterns, guardrails, and quality expectations that guide both human squads and AI agents.
The Chapters and CoCs maintain capability excellence across product, design, engineering, data and AI, cloud, security, QA, Agile, DevOps, architecture, and industry.
The Portfolio Teams manage business-aligned portfolios and sub-portfolios, with sticky squads that build domain knowledge over time.
The Co-create Squads remain human-led and AI-accelerated, focused on strategy, value framing, discovery, pain point validation, solution envisioning, and business opportunity development.
The Co-execute Squads become specification-driven AI execution pods, focused on translating business intent into working solutions through AI-assisted design, engineering, testing, and release.
The Co-operate Squads become agentic operations pods, focused on reliability, adoption, continuous improvement, value realization, and scaled operational governance.
The model does not remove humans.
It moves humans to the highest-value work.
How to Move from the Current Model to the Future Model
The transition should not be treated as a big-bang reorganization. It should be a deliberate evolution.
Step 1: Map the Current Operating Model
Start by documenting the current portfolios, sub-portfolios, squads, boards, chapters, CoCs, ceremonies, governance flows, value tracking mechanisms, and escalation paths.
Identify where the current model already works well. Do not change those areas unnecessarily.
The goal is not to replace the Garage operating model. The goal is to make it agentic.
Step 2: Define the Agentic North Star
Create a clear statement of what the future model is meant to achieve.
For example:
“We will evolve IBM Garage into a human-led, AI-accelerated operating model where small, sticky squads use AI agents to move from strategy to value realization faster, while strengthening governance, standards, adoption, and measurable business outcomes.”
This north star should make clear that AI is not replacing the operating model. AI is accelerating it.
Step 3: Redesign Squad Archetypes
Define the new squad archetypes for Co-create, Co-execute, and Co-operate.
The default archetype should be the three-person AI-augmented squad:
- Product Owner
- Forward Deployment Engineer
- Designer / Design Researcher
Then define when additional roles are needed, such as change manager, security specialist, data engineer, architect, SRE, QA specialist, or industry SME.
This creates flexibility without reverting to large default squads.
Step 4: Establish Sticky Pods
Move away from treating people as interchangeable resources.
Create sticky pods aligned to portfolios or sub-portfolios. Allow these pods to move from initiative to initiative while preserving domain knowledge, stakeholder context, technical familiarity, and delivery rhythm.
This is one of the most important operating model changes.
AI can generate output quickly, but sticky human context creates better judgment.
Step 5: Upgrade the Design Authority
Expand the Design Authority mandate from reviewing work to defining the standards that guide agentic work.
This includes reusable patterns, prompt libraries, solution templates, reference architectures, design systems, security policies, testing standards, documentation standards, and responsible AI controls.
Then introduce AI agents that perform first-level reviews against those standards.
The Design Authority becomes both a governance body and an enablement engine.
Step 6: Strengthen the Value Office
Define value orchestration as a core capability, not an optional reporting function.
Every initiative should have clear value drivers, value levers, KPIs, baselines, targets, ownership, and realization timing.
The Value Office should track whether initiatives are producing the expected value and should escalate when they are not.
This prevents the organization from scaling AI activity without scaling business impact.
Step 7: Rewire Ceremonies for AI-Augmented Work
Keep the ceremonies, but change the inputs and outputs.
Product increment planning should use AI-generated story maps, test plans, dependency maps, and delivery options.
Portfolio sync should use AI-generated portfolio health insights, value signals, risk summaries, and capacity scenarios.
Architecture review should use AI-generated standards checks.
Impediment management should use AI-generated blocker summaries and escalation recommendations.
The ceremony stays human. The preparation becomes AI-assisted.
Step 8: Pilot the Model in One Portfolio
Do not transform everything at once.
Select one portfolio with meaningful business demand, supportive leadership, available SMEs, and enough delivery complexity to test the model properly.
Run the new model across one or two business opportunities. Measure speed, quality, adoption, team experience, value realization, and governance effectiveness.
The goal is to learn what changes when squads become smaller, AI usage increases, and governance becomes more standards-driven.
Step 9: Build the Agentic Enablement Layer
Create the reusable assets that allow the model to scale.
This includes:
- Standard prompts
- Agent workflows
- Specification templates
- Architecture review checklists
- Design review checklists
- Business opportunity canvases
- Value framework templates
- Testing patterns
- Security review patterns
- Delivery playbooks
- Adoption playbooks
- Governance dashboards
Without this enablement layer, each squad will invent its own way of working. That creates inconsistency and risk.
Step 10: Scale Across Portfolios
Once the model proves itself, scale it across portfolios.
Use the Garage Board, Design Authority, Value Office, chapters, and CoCs to ensure consistency. Use the sticky pod model to preserve knowledge. Use the Investment Board to make sure funding follows value.
The future operating model should scale through patterns, not bureaucracy.
The Real Shift: From Delivery Capacity to Decision Quality
The biggest change in the future Garage operating model is not that AI writes more code or produces more documentation.
The biggest change is that the bottleneck moves.
In the old model, the bottleneck was often delivery capacity.
In the future model, the bottleneck becomes decision quality.
Can the organization decide which opportunities matter most? Can it define value clearly? Can it write good specifications? Can it set strong standards? Can it manage risk? Can it create adoption? Can it govern AI-generated work without slowing everything down? Can it stop low-value work quickly? Can it scale what works?
That is why the IBM Garage operating model remains so relevant.
Garage was never just a delivery method. It was always a way to connect business intent, user needs, technology execution, and measurable value.
Agentic AI does not make that obsolete.
It makes it essential.
Conclusion: The Garage Becomes the Operating System for Agentic Transformation
The future of IBM Garage is not a choice between humans and AI.
It is a model where humans and AI work together through a disciplined operating system.
Humans set direction. AI accelerates work.
Humans define value. AI generates options.
Humans make decisions. AI provides evidence.
Humans own adoption. AI supports change.
Humans govern risk. AI scales review.
Humans orchestrate outcomes. AI compresses the path from idea to impact.
Organizations that adopt IBM Garage should not simply add AI tools to existing squads and call that transformation. They should evolve the operating model itself.
The future Garage is portfolio-based, value-led, governance-enabled, AI-accelerated, and human-accountable.
It keeps the boards, chapters, ceremonies, portfolios, and value discipline that enterprises need.
It transforms the squad into a small, sticky, AI-augmented pod.
It turns Co-create into a faster but still human-led discovery and envisioning engine.
It turns Co-execute into a specification-driven agentic delivery system.
It turns Co-operate into an AI-enabled value realization and operational excellence model.
That is the next evolution of IBM Garage.
Not less Garage.
A more powerful Garage for the agentic enterprise.
