State of AgenticInfrastructure 2026
01
Who we surveyed
Key insights
- Our respondents are hands-on practitioners: engineers spread across product, DevOps, platform, and ops teams rather than a single infrastructure group.
- The sample centers on mid-sized software companies, not startups or large enterprises, but teams with real scale that are still moving fast.
By the numbers
- 510 platform, DevOps, and product engineers
- Product and app developers are the largest group (35%), ahead of DevOps (23%) and platform engineering (17%)
- 65% work at companies of 201–5,000 people, and 65% are in Software / SaaS
Takeaway
Among the people we surveyed, infrastructure work reaches well beyond the infra team: a third sit on product or app development teams.
01.Q1
Which best describes your team?
01.Q2
How large is your company?
01.Q3
What industry are you in?
02
The infrastructure baseline
Key insights
- These teams are mature and ship fast: quick deploy cadence and self-service platforms are already standard.
- But governance stays manual: review gates (61%) still outweigh policy-as-code in CI (54%).
By the numbers
- 82% deploy to production weekly or more often
- Multi-cloud is the norm: AWS 64%, Google Cloud 55%, Azure 45%
- 63% already run an internal developer platform or golden paths in production
Takeaway
Speed is solved; control isn’t. Only 7% cite slow provisioning. The real pain is consistency (28%), cost (24%), and security (20%).
02.Q5
Which cloud providers does your infrastructure run on today?
02.Q6
How often do you deploy infrastructure changes to production?
02.Q7
How do you handle policy and compliance in your infra pipeline?
02.Q8
What's your biggest technical challenge in managing infrastructure?
02.Q9
How is infrastructure ownership structured on your team?
- Dedicated platform/infra team 47%
- Embedded in product teams 30%
- Fully self-service for developers 22%
- No clear ownership 2%
02.Q10
Do you have an internal developer platform or golden paths for provisioning?
- Yes, in production 63%
- Building one now 24%
- No plans 14%
03
Agents in the workflow today
Key insights
- The headline tools split by size: Claude Code is the favorite at companies under ~200, while Copilot’s lead grows with headcount, reaching 70%+ in orgs of 1,000 or more.
- Today’s agents mostly review and advise rather than write; authoring infrastructure code trails every other use at 29%.
- 81% let agents change production, but almost all of that is gated: “with approval” (62%) far outweighs “autonomously” (19%).
By the numbers
- Only 4% use no AI in their infrastructure workflow
- AI shows up most in code review (70%), security scanning (56%), and cost optimization (52%)
- GitHub Copilot (62%) and Claude Code (56%) lead the tools in use
- 45% say agents already handle half or more of their infra work
Takeaway
Agents are everywhere in the workflow, but a human still signs off before production.
03.Q11
Where are you using AI or agents in your infra workflow today?
03.Q12
Which AI tools do you use for infrastructure work?
03.Q13
How much of your infrastructure work is done by agents?
03.Q15
Do you let AI agents change production infrastructure?
- Yes, autonomously 19%
- Yes, with approval 62%
- Only in non-prod 9%
- No 10%
04
AI in monitoring and operations
Key insights
- Monitoring is one of the most widely adopted places for AI, led by anomaly detection.
- Autonomy stays gated even here: “suggests fixes” (37%) and “remediates with approval” (31%) are the common modes.
By the numbers
- 64% already use AI for infrastructure monitoring; just 6% have ruled it out
- Top uses are anomaly detection (57%), auto-remediation (45%), and predictive scaling (44%)
- Just 12% run fully autonomous monitoring
Takeaway
Even the most-adopted use case keeps a human in the loop: agents suggest, people approve.
04.Q16
Do you use AI for infrastructure monitoring?
- Yes 64%
- Evaluating 31%
- No 6%
04.Q17
What do you use AI-driven monitoring for?
04.Q18
How autonomous is your AI monitoring?
05
The six-month outlook and sentiment
Key insights
- Teams expect to hand agents more work: across the board, the share of infrastructure they expect agents to generate rises over the next six months.
- 63% say they trust agents to make production changes, which runs ahead of the manual-approval reality from the earlier sections.
By the numbers
- Teams expecting agents to generate 50% or more of their infra code rise from 45% today to 52% in six months
- The “under 25%” group shrinks from 23% to 15%
- 82% agree AI will meaningfully change how they write infrastructure within 12 months
- 72% say proactive AI monitoring has reduced their incident volume
Takeaway
Stated trust is outrunning real guardrails: 63% say they trust agents in production, yet almost every change still needs manual approval. The next six months are about closing that gap.
05.Q13 → Q14
Share of infrastructure code that is AI-generated, today vs. in six months
- Today
- In six months
05.Q19
How strongly do teams agree?
- Strongly disagree
- Disagree
- Neutral
- Agree
- Strongly agree
AI and agents will meaningfully change how we write infrastructure within 12 months
Our current tooling makes it easy to onboard a new engineer to our infrastructure
Managed platforms have reduced our operational burden
Proactive AI monitoring has reduced our incident volume
I trust AI agents to make production infrastructure changes
We spend too much time on undifferentiated infrastructure toil
Put agents to work on your infrastructure
An AI agent that plans and executes infrastructure changes, and waits for your approval before touching production.
Explore NeoDefine infrastructure in TypeScript, Python, Go, and more: the real languages agents already know how to write.
Explore IaCConnect Claude Code, Copilot, and your own agents to your infrastructure through the Model Context Protocol.
Read the docsKnowledge packages that teach Claude Code, Cursor, and other agents to build and manage infrastructure.
Explore skills