Invention Discovery in AI-Augmented Sprints

Developers using GitHub Copilot complete tasks 55.8% faster than those without it, per Microsoft Research. The productivity gains are real. So is the cost: when sprints compress from weeks to days, documentation shrinks with them. Novel solutions get committed, merged, and forgotten. Features that would have been flagged as inventive in a slower cycle disappear into the backlog.

AI-related patent grants grew from 34,544 in 2020 to 54,022 in 2024, per Anaqua's analysis of USPTO data. Competitors are filing. Teams that skip invention logging during development are leaving patentable work uncaptured.

Last updated: March 2026. This page is informational only and not legal advice. Consult a patent attorney for your specific situation.

55.8%

Faster task completion with GitHub Copilot, per Microsoft Research

54,022

AI-related patent grants in 2024, up from 34,544 in 2020 (Anaqua/USPTO)

>20%

Of U.S. organizations patenting in AI by 2018, up from under 5% in 1980

12 mo

Window to file a non-provisional after a provisional to claim priority

Why Agile Sprints Leak IP

Agile was built for shipping software, not capturing inventions. The standard 2-to-4 week sprint cycle rewards velocity, working demos, and iterative delivery. None of those incentives overlap with patent disclosure.

Several failure points show up consistently across engineering teams:

Planning meetings focus on user stories, not novelty. Stories are scoped for delivery. Nobody asks whether the proposed solution is new to the world.

Code reviews check for bugs, not inventive steps. Reviewers flag logic errors and style issues. Inventive architecture passes through unlogged.

Retrospectives evaluate process, not cumulative innovation. Teams reflect on what slowed delivery. Not on what novel methods they developed along the way.

External demos may trigger prior-art or on-sale bar concerns. Demos shown to users, investors, or partners without confidentiality agreements may create disclosure risk under 35 U.S.C. § 102.

AI-generated code creates a documentation gap. When Copilot or a similar tool generates a solution, developers may have conceived the architecture, directed the implementation, and validated the result. If nobody logs that contribution, there is no record to support inventorship later.

Key point: Human conception and contribution remain the legal standard for inventorship. AI tools do not disqualify you. Undocumented contributions can.

A Sprint-Level Invention Logging Workflow

You do not need a separate IP process. You need invention checkpoints embedded into the sprint cycle you already run. Here is a four-stage approach.

Sprint planning. When the team scopes new features, add one question to every story: "Does this solve a problem in a way we have not seen before?" Flag stories that get a yes. Two minutes per story. That is your initial discovery queue.

During development. Attach a structured disclosure template to flagged stories. A shared doc with fields for problem, solution, prior art awareness, and human contribution works. Developers fill it in as they code. In Jira or Linear, a custom field or linked page does the job. Capture the inventive step while context is fresh.

Before demo or external review. Any feature shown to users, investors, or partners outside a confidentiality agreement may create disclosure risk under 35 U.S.C. § 102. Consider filing a provisional before external review when the feature may be patentable and disclosure is imminent. Check the current USPTO fee schedule for applicable filing fees, which vary by entity size.

Sprint retrospective. Dedicate 10 minutes to reviewing the disclosure queue. Some sprints produce nothing patentable. Others reveal cumulative innovations across several iterations that only become visible in hindsight. A later non-provisional can claim priority to an earlier provisional, but only to the extent the provisional adequately supports the claimed subject matter.

Remember: Multiple provisionals do not automatically combine. The non-provisional must be drafted to capture and support the specific claims intended to rely on each earlier filing.

Inventorship When AI Writes the Code

Under 35 U.S.C. § 100(f), an inventor must be a natural person. AI cannot be listed as an inventor. The USPTO has addressed this directly: the patent system was developed to incentivize human ingenuity, and AI systems cannot be named as inventors. That guidance has been refined over time, and borderline cases remain fact-specific.

The Legal Standard

Informed by Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998), each named inventor must make a significant contribution to conception. Using AI tools does not disqualify you. You must be able to show meaningful human contribution.

What Counts as Human Contribution

Directing the AI, selecting among outputs, combining solutions, identifying the problem the AI helped solve, and validating or modifying generated code can all support inventorship. Document each of these choices as you work.

What Your Disclosure Template Needs

Add a dedicated field: What did the developer conceive? What did they direct the AI to do? What judgment calls did they make? This becomes your evidence of inventorship if the patent is ever challenged.

Borderline Cases

When the line between human direction and AI generation is unclear, consult patent counsel before filing. Later developments have not eliminated uncertainty around edge cases involving substantial AI contribution.

Not Everything Deserves a Patent

Systematic logging does not mean systematic filing. Patents are expensive. Software patent prosecution costs are substantial, and eligibility challenges under Alice Corp. v. CLS Bank make software claims harder to secure.

Some innovations are better protected as trade secrets. Proprietary algorithms, training data pipelines, and internal tooling may never need a patent if you can keep them confidential. The key is making that decision deliberately, not by default because nobody noticed the invention.

Your sprint retrospective is the right place for triage. Sort disclosed innovations into three buckets:

File provisional. Novel methods with a clear path to a defensible patent claim, especially where competitors could independently discover or reverse-engineer the approach.

Protect as trade secret. Algorithms, pipelines, and internal tooling that can remain confidential and provide ongoing competitive advantage without public disclosure.

Release. Implementations with low novelty, high prior-art risk, or where open publication supports business goals: developer community, recruiting, thought leadership.

Practical tip: Local-first patent scanning tools can support triage by surfacing potential overlaps with existing patents or flagging novel methods, without requiring source code to leave your environment.

Building the Culture That Makes This Work

Workflows fail without incentives. If your engineers see invention logging as overhead, they will skip it. A few approaches tend to work for early-stage teams.

Onboarding Context

Make IP awareness part of onboarding. A 30-minute session on what counts as patentable, how provisionals work, and why documentation matters establishes baseline context before any sprint begins.

Recognition Incentives

Tie invention disclosures to recognition. Some companies offer bonuses per filed provisional. Others use public recognition in team channels. Either approach signals that the company values discovery alongside delivery.

Confidentiality Agreements

Use NDAs for external reviews. Every demo, beta test, and investor meeting is a potential disclosure event. NDAs reduce that risk, but they are not a substitute for timely filing when a feature may be patentable.

Long-Term IP Records

The share of U.S. organizations patenting in AI grew from under 5% in 1980 to over 20% in 2018 (USPTO). Teams that build logging habits early accumulate more complete IP records. That matters during due diligence and licensing discussions.

This page is for informational purposes only and does not constitute legal advice. Patent strategy depends on jurisdiction-specific rules and the facts of each situation. Consult qualified legal counsel for guidance on your specific circumstances.

Frequently Asked Questions

Can AI be named as an inventor on a patent?

No. Under 35 U.S.C. § 100(f), only natural persons qualify as inventors. The USPTO has addressed this in guidance, and subsequent revisions have not changed the rule. Your developers can use AI tools freely, but they must demonstrate significant human contribution to the conception of any claimed invention.

How much does a provisional patent application cost?

USPTO filing fees for provisional applications vary by entity size and are updated periodically. Check the USPTO fee schedule directly for current amounts. Attorney preparation costs vary as well. A provisional requires less formality than a non-provisional and establishes a priority date while you decide whether to pursue a full application within 12 months.

Does showing a feature in a sprint demo count as public disclosure?

It depends on the audience and the conditions of the disclosure. A demo to your internal team under existing employment agreements is generally not a public disclosure. A demo to external users, customers, or investors involves more risk. The analysis turns on whether the disclosure was publicly accessible and sufficiently enabling. Some external demos may trigger prior-art or on-sale bar concerns under 35 U.S.C. § 102. File a provisional or use confidentiality agreements before external reviews when the feature may be patentable.

Should we patent everything our team invents?

No. Patent prosecution is expensive, and software patents face eligibility hurdles under Alice Corp. v. CLS Bank. Some innovations are better protected as trade secrets. The goal of systematic logging is to make that decision deliberately during sprint retrospectives, not to file on everything automatically.

How do we document human contribution when using AI coding tools?

Add a human contribution field to your invention disclosure template. Developers should describe what problem they identified, how they directed the AI tool, what design choices they made, and how they validated or modified the output. This record supports inventorship under the Pannu v. Iolab Corp. factors if the patent is ever challenged.

What if we miss a patentable invention in a past sprint?

U.S. patent law includes a grace period under 35 U.S.C. § 102(b)(1) that may apply to certain prior disclosures made by the inventor, but its scope is limited and fact-dependent. It does not cover all types of earlier disclosures, and prior public disclosures can still complicate patentability. Review past sprint artifacts and commit histories, then consult counsel to assess whether filing is still viable.

How many inventions slipped through your last sprint?

ObviouslyNot scans your codebase on-device to surface distinctive methods before they disappear into the backlog. Integrate it into your retrospective in minutes.

No code leaves your device.

Scan your codebase free