Sense and Respond

When amplification happens well, signals are generated and transmitted so that they are received and understood clearly by the receiver and the appropriate corrective actions can be taken. This requires generating and transmitting signals that are frequent enough, fast enough, precise enough, accurate enough, and loud enough to get us to decide to act differently based on what is being observed and understood.

— Gene Kim, Steven J Spear [1]

What is the Sense and Respond event? 

SAFe has always used nested cadences – iterations, typically a week or two weeks long, and planning intervals (PIs) every 8 to 12 weeks. While the nature and purpose of the cadences remain fairly consistent, the increased variability and speed introduced by AI have driven significant changes to the events associated with them.

In AI-Native SAFe:

• The half-day Inspect and Adapt event that drove ART-level learning cycles on a PI cadence has been replaced by a two-hour Sense and Respond event every iteration, driving accelerated learning. This Sense and Respond event also includes the activities formally present in the System Demo.

NOTE: Some ARTs may choose to run the problem-solving portion of the Core SAFe Inspect and Adapt event periodically if it supports their ongoing approach to problem-solving.

• The two-day PI Planning event that provided the time and space for the entire ART to create a set of plans for the upcoming PI has been replaced by a one-day PI Outcome Planning event that focuses on in-flight re-alignment.

These events are shown in Figure 1 below:

The Sense and Respond event is an AI-Native ART event designed to translate evidence and data into actionable decisions. Sense and Respond events occur on a cadence, at least once every iteration. These events allow the ART to pause, evaluate the rapid flow of local experiments, and ensure that collective learning is applied across the entire system.

While AI-Native Teams take a flow-based approach, operating at high speed with high levels of autonomy, the ART Sense and Respond event provides the necessary cadence to maintain technical coherence and strategic alignment across all teams.

What is the purpose of the Sense and Respond event?

Sense and Respond events, operating on a cadence, provide a critical mechanism for achieving three things:

1. Managing the variability inherent in AI product development

Variability is inherent in product development, and never more so than in the Age of AI. The context in which our products need to succeed, the AI models they’re based on, and the customer needs we’re attempting to satisfy are all changing more rapidly than ever before.

As teams sense and respond independently, they begin to diverge on several fronts at once:

• Insights – some teams sense signals that other teams have yet to notice
• Interpretation – the same feedback is understood in different ways
• Intent – teams adapt in directions that diverge from how others understood them

Whilst effective use of team-level align-sense-respond cycles enables AI-Native Teams to adapt to this level of variability, that local adaptation magnifies the level of variability at play when developing complex products involving multiple teams. 

This divergence compounds over time. Divergent insights lead to divergent interpretations, which in turn lead to divergence over the product’s intent. Left unchecked, this variability continues to accumulate faster with every turn of the learning loop. 

2. Intentionally discarding work that doesn’t drive outcomes

This variability is compounded by the fact that AI enables teams to greatly increase their output. Specifically, allowing teams to ‘overproduce’ as they seek to uncover the set of features most impactful to customers. This approach to innovation, however, only works if there is discipline to stop pursuing areas of exploration that fail to drive outcomes. The Sense and Respond event provides this forcing function.

3. Sharing learning across the ART

AI presents opportunities for teams to continually uncover new ways of working and new ways to build the product itself. The continuous learning environment is hugely motivating and exciting for teams that are using technology to push on the boundaries of what is possible. However, this learning can quickly become isolated. The teams are naturally focused on what is most important for them. Leaving little or no time for sharing across the entire ART. The Sense and Respond event provides the time and space for this ART-level systemic learning.

How to run a Sense and Respond Event

How often does the Sense and Respond Event occur?

Sense and Respond events occur throughout the PI, at a minimum at iteration boundaries. Additional Sense and Respond events can be triggered immediately by specific circumstances or significant signals that require an immediate system-level response. No important signal should wait until the end of an iteration or PI to be measured or addressed.

Who attends the Sense and Respond event?

The following attendance ensures the event has the necessary judgment, context, and domain expertise to evaluate AI-generated evidence.

• ART Leadership: Product Management, System Architects, and Release Train Engineers.
• Stakeholders: Business Owners who monitor economic evidence and risk, and other stakeholders closely associated with the work of the ART.
• Experts: AI Value Architects who coach the ART on responsible AI adoption and workflow improvements.
• AI-Native Teams: Ideally, all team members prioritize attendance. At a minimum, team members who are needed to provide context on the outcomes and experiments that will be discussed in detail.

Every member of the ART adds value to the conversations during the Sense and Respond event. Prioritizing attendance and ensuring engagement during the event ignites innovation in everyone. This spark is what creates great ARTs.

What is the structure of a Sense and Respond event?

The Sense and Respond event is a two-hour cadence-based event in two parts that ensures it remains effective for decision-making.

Part 1: Sensing Activities

The first half of the event focuses on shared sensing, ensuring insights, feedback, and data are heard and understood by everyone in attendance. Four specific sensing activities are identified

1. Progress against Outcomes

The current PI and Team outcomes are presented, along with the updated key results, to demonstrate progress against them.

Business Owners, Product Management, or other stakeholders may also want to share any significant strategic shifts since the last Sense and Respond event, and how they directly affect the outcomes the ART is pursuing.

2. Product Demos and Insights

Teams demonstrate completed features, enablers, experiments, and prototypes, and share the measurable results achieved. Importantly, not every AI-Native Team has to demo every Sense and Respond event. RTEs should work with teams to identify which items are most crucial for being demoed

Examples might include:

• Completed work that directly influences outcomes at the team, PI, or ART level.
• Work that deserves to be showcased to recognize the significant efforts from a team and provide inspiration to the ART.
• Experiments, prototypes, or learnings that could impact other teams, foster collaboration, and knowledge sharing.
• Communicating completed initiatives to ensure everyone interacting with customers has the right messaging.

3. Workflow Improvements and Challenges

Teams showcase workflow improvements and other AI-driven innovations.The RTEs and Value Architects can help by working with teams to identify improvements most likely to be leveraged by other teams or scaled and incorporated into the Continuous Innovation and Delivery Pipeline (CIDP).

Ensure opportunities with significant promise are considered, even if they are still in a highly experimental state. Extending experimental innovations across multiple teams can enhance their impact and resilience long before reaching the level of proof required for formal incorporation in the CIPD.

Additionally, teams may often simply present challenges that they are facing directly. These challenges can be varied. They might include challenges connected to technology, bottlenecks that are slowing them down, access to required data, etc. The chances are that another team on the ART has simply challenged and solved it, or someone at the event is able to volunteer to work with the team to address it.

4. Customer Feedback and Metrics

Review evidence from multiple input sources. Common examples include, but are not limited to:

• Product performance measures.
• Operational measures, such as time to market and flow efficiency.
• Qualitative and quantitative customer feedback, including feedback gathered from customer demos.
• AI signals such as governance and risk health, model performance, system drift, and hallucinations.

Part 2: Responding Activities

The second half of the event is dedicated to aligning on the set of responses that will ensure the ART successfully achieves its outcomes. Three response activities should be included as described below.

1. Identify Responses

While presenting and discussing sensing activities in the first half of the event, it is common to track the items most likely to require a response. This is the time to review that list, add anything additional to it, and agree on the items that may require a decision or action.

2. Explore Options

Different approaches can be used for this activity. What is important is taking the time to explore the available options. This prevents falling into the trap of taking the most obvious response and also creates space for multiple voices to be heard.

Consider an example.

In this Sense and Respond event, one team shared the results of some experiments it had been running to evaluate different approaches to integrating an AI-driven customer recommendation engine. The experiments demonstrated that one path was immediately truncated, but three alternatives remained. Each one scored highly across different evaluation criteria. One was highly user-friendly, another highly performing, and the third easily integrated into existing systems.

In this example, a response is clearly needed. It doesn’t make economic sense for the team to continue experimenting at current levels. But before a response is determined, the options should be considered. One option is to select from the three available choices. Another option could be to explore whether the benefits of the three experiments could be combined with some further experimentation. Another option might be for the team to evaluate against current security policies to see if that naturally reduces the possibilities.

Facilitating this activity can be done in different ways. If it is a topic that broadly affects the ART, the RTE may choose to run it as a combined discussion. However, even in these situations, it can be useful to create smaller breakouts where options can be discussed more easily before returning to share findings.

3. Agree Decisions and Actions:

Once the options are available for all the items that need a response, it is time to process them.

For some, the decision or action will be clear, and it can be taken immediately at the time of the event, with a clear owner assigned and next steps determined as appropriate. In other situations, more time and thought are required, along with further investigation. The actions that describe this additional work should still be agreed upon.

Be mindful not to defer all decisions and actions until after the event. The next Sense and Respond event is just around the corner. Getting into the habit of more regular, incremental decisions is key to navigating the opportunities and challenges that AI presents.

How should the Sense and Respond Event be facilitated? 

The Release Train Engineer (RTE) facilitates Sense and Respond events. While facilitating the event, the RTE should have three perspectives in mind, which can be illustrated using three key questions:

1. Objective Outcome Progress: Are we evaluating progress against ART and PI outcomes using objective data?
2. Listening to Customer Feedback: Are we reviewing how outputs have been received by customers using product performance metrics and direct feedback from customer demos?
3. Prioritizing Workflow Improvements: Are we identifying workflows that can be optimized based on evidence, and are these the most important workflows to improve?

Additionally, the RTE must ensure that the event avoids becoming a status report and instead focuses on “sensing” the cumulative impact of recent work and “responding” with decisions that keep the ART aligned with its vision.

What is an example Sense and Respond event agenda?

The agenda created from these items will vary based on how much time each ART wants to spend on them. An example agenda with a typical weighting across these activities is shown in Figure 2 below.

Sharing learning after the event

Creating the opportunity for shared learning is central to the Sense and Respond event. During the Sense and Respond event, tacit learning is taking place. This is learning that moves through human conversation and collaboration during events, as well as through individual and team interactions.

However, this is only part of the learning process, which also includes ‘encodable’ and ‘contextual’ learning as shown in the examples below:

Following the event, the RTE must ensure that the actions and decisions to update the systems that manage and maintain our encodable and contextual learning are also carried out. Ideally, the necessary updates should be clear from the decisions and actions taken during the Sense and Respond event. Over time, it will become second nature to ask questions during the event, such as ‘what is the impact of this decision on our contextual learning?’ 

Ensuring these three types of learning are addressed is a critical step that ultimately strengthens the ART and ensures it can carry out its work more effectively.

---

References

[1] Kim, Gene; Spear, Steven J. Kim, Gene, and Steven J. Spear. Wiring the Winning Organization: Liberating Our Collective Greatness through Slowification, Simplification, and Amplification. Portland: IT Revolution, 2023.