Agile Spike
What is an Agile Spike?
An agile spike is a time-boxed task meant to answer a question or reduce uncertainty so the team can estimate or design upcoming work. It produces knowledge, such as a decision or throwaway prototype, rather than a shippable feature.
Updated November 2026 · 3 min read
An agile spike is a time-boxed piece of work whose purpose is to answer a question or reduce uncertainty, not to deliver a feature. A team uses a spike when it cannot estimate or design a backlog item because something important is unknown, such as whether a third-party API can handle the load, which of two technical approaches is simpler or how users will respond to a new workflow. The output is knowledge: a decision, a prototype to throw away, an estimate the team trusts or a set of better-written stories.
The term comes from Extreme Programming, where a spike was a quick, narrow experiment, often a small piece of throwaway code, built to explore a technical problem. It is now used across agile frameworks, including Scrum and SAFe.
Types of spikes
| Type | Question it answers | Example |
|---|---|---|
| Technical spike | Can we build it, and how? | Testing whether a search library returns results fast enough on production-sized data |
| Functional spike | What should it do, and how will users interact with it? | Prototyping two checkout flows and testing them with five customers |
| Estimation spike | How big is this, really? | Exploring a legacy module to understand how much would need to change |
SAFe formally distinguishes technical and functional spikes; the estimation spike is a common informal label for a spike whose main aim is a reliable size.
When to use a spike
- The team cannot agree on an estimate because of a specific unknown.
- Two or more technical approaches look viable and choosing the wrong one would be expensive.
- A story depends on an external system, vendor or dataset nobody on the team has used.
- A risk, such as performance, security or regulatory interpretation, needs checking before committing.
If the unknown is small enough to resolve in an hour of conversation, a spike is overkill.
How to run a spike
- Write the question. State exactly what the team needs to learn, such as "Can the payment provider's API return a refund status within two seconds at our peak volume?"
- Define the output. A decision, a recommendation, a revised estimate or new stories. Agree it in advance.
- Time-box it. Usually a few hours to a few days, and no more than one sprint. When time runs out, the team reports what it learned even if the answer is incomplete.
- Put it on the backlog. Add the spike as a backlog item, ordered by the product owner like any other work, so it is visible and planned.
- Share the result. Present findings at refinement or review, and update the stories that depended on it.
- Discard throwaway code. Prototype code written for learning should not slip into production without being rebuilt to standard.
An example
Consider a hypothetical team asked to add address lookup to a signup form. Estimates for the story range from 3 to 13 story points, because nobody knows whether the shortlisted address data provider covers rural addresses well. The product owner adds a spike, time-boxed to one developer for two days, with the question "Does provider A return a valid match for at least 95% of a sample of our real customer addresses?"
The developer tests 1,000 anonymized addresses from the customer database. Provider A matches 962, a rate of 96.2%, but misses most addresses in two rural regions. The team decides to use provider A with manual entry as a fallback, writes two stories (lookup, and fallback entry) and estimates them at 5 and 2 points. The two-day spike replaced a guess that ranged from 3 to 13 points with an estimate of 7 that the team agrees on.
Common pitfalls
- No time-box. Open-ended research tends to expand until it becomes a project.
- Vague questions. "Investigate the new API" produces notes, not decisions.
- Too many spikes. A sprint full of spikes delivers no value. Frequent spikes can signal that the team needs more discovery earlier.
- Counting spikes as velocity. Many teams do not assign story points to spikes, or track them separately, so velocity reflects delivered value.
- Shipping the prototype. Exploratory code usually lacks tests and error handling.
Spikes help teams split work into well-understood user stories, make estimation more reliable and bring items to the team's definition of ready during backlog refinement.
Frequently asked questions
What are the types of spikes in agile?
A technical spike asks whether and how something can be built, such as testing a search library's speed on production-sized data. A functional spike asks what a feature should do and how users will interact with it, for example by prototyping two flows. An estimation spike, an informal label, explores unknown work such as a legacy module to produce a reliable size.
How long should an agile spike last?
An agile spike should be time-boxed, usually from a few hours to a few days and never more than one sprint. When the time runs out, the team reports what it learned even if the answer is incomplete. Open-ended research tends to grow into a project, so agree the question and the expected output, such as a decision or revised estimate, before starting.
Should agile spikes be given story points?
Many teams do not assign story points to spikes, or track them separately, so that velocity reflects delivered value rather than research. Spikes should still go on the backlog and be ordered by the product owner like any other work, which keeps them visible and planned. A sprint filled with spikes may signal that more discovery is needed earlier.