What 69,000 Jira Tickets Say About AI in Engineering

LinkedIn Carousel — export instructions

Print to PDF: Click the button below → in the print dialog choose "Save as PDF", paper size will be set to 1080×1080px per slide.
Or screenshot each slide: Right-click each slide → "Save image as" using a browser extension like GoFullPage, or screenshot each one individually.
Upload to LinkedIn: Create a post → click the document icon → upload the PDF. LinkedIn will show it as a swipeable carousel.
Post tip: Put the link to task2vec.com in the first comment, not the post body — LinkedIn suppresses links in posts.

Post text (copy before uploading)

Paste into the LinkedIn post body. Put the link in the first comment — LinkedIn suppresses links in post text.

AI won't make your engineering team smaller.

It will make it harder to staff.

We analysed 69,000 Jira tickets and 11,000 Git commits from the Spring Framework — one of the world's most-used Java projects, 20 years of history.

In 2005, 39% of the work was routine enough to safely delegate to AI.

By 2020: 5%.

The easy problems are gone. What remains is architecture decisions, concurrency bugs, and cross-cutting design questions that need hard-won experience.

That creates two problems most engineering leaders haven't thought through:

1. You need more senior engineers, not fewer
2. You no longer know how to grow them — because the apprenticeship path (start with easy tickets, build up to hard ones) is quietly disappearing

Swipe to see the 15-year trend →

Next post: what does an expert-in-training actually look like in the ticket data?

#EngineeringLeadership #SoftwareEngineering #AI #Jira #DeveloperProductivity

First comment text

Full research + interactive explorer: task2vec.com/research
Commission a study on your own data: task2vec.com/commission

Slide 1 / 6 — Cover

task2vec · 2025

What 69,000
Jira tickets say
about AI.

69,156 Jira tickets + 11,000 Git commits.
Here's what the data says about AI in software engineering.

1 / 6

Slide 2 / 6 — Three tiers

Spring Framework · 69,156 Jira tickets + 11k Git commits · 2002–2021

22%

Automate

Safe for AI with no human review — docs, deprecations, typo fixes, dependency bumps

30%

Assist

AI draft + human sign-off — moderate complexity, some context required

48%

Escalate

Senior engineer required — architecture decisions, concurrency, failure modes

2 / 6

Slide 3 / 6 — The trend

How it changed over time (2005–2020)

From 39% to 5%
in 15 years.

2005

39% auto

45% esc

early peak

2010

19% auto

36% escalate

routine work declining

2015

19% auto

49% escalate

plateau then cliff

2018

15% auto

59% escalate

2020

5% auto

68% escalate

only hard problems remain

Automatable

Escalate

3 / 6

Slide 4 / 6 — Key insight

Key insight

AI doesn't reduce
the hard pile.
It exposes it faster.

In 2005, 39% of Spring Framework tickets were routine enough for AI. By 2020 that had fallen to 5%. The routine work got done — and what's left is architecture decisions, failure modes, and cross-cutting concerns that genuinely require senior engineers.

The triage tool becomes more valuable as adoption grows, not less.

4 / 6

Slide 5 / 6 — The implication

Can AI help our team?

Which 22% of your backlog
is safe to automate?

Every team has a different split

Spring's 22/30/48 is from an open-source framework. Your product codebase will be different — possibly much more automatable if it's younger.

Getting it wrong is expensive

An AI-drafted PR for an escalate-level ticket doesn't save time. It creates review burden, subtle bugs, and erodes trust in the tooling.

Triage first. Automate second.

The same methodology we used on Spring can be run on your own Jira history.

5 / 6

Slide 6 / 6 — CTA

task2vec

See the full
analysis.

Open research · Peer-reviewed methodology · All data public

Results & model comparison

task2vec.com/eval_summary

→

Run it on your own Jira

task2vec.com/commission

→

Link in first comment · Code open source at github.com/JussiTu/task2vec

6 / 6