Briefs · III · Ideatives Inc. · Updated

Production is the test.

36% of respondents say their organization has a high-impact outage weekly or more often, according to New Relic's 2026 Observability Forecast (2,575 respondents in 24 countries; vendor-funded).

New Relic 2026 Observability Forecast · 2,575 respondents, 24 countries, vendor-funded

How often do companies have high-impact outages?

In New Relic's 2026 Observability Forecast (2,575 respondents, vendor-funded), 36.2% reported high-impact outages weekly or more, 46.7% two to three times a month or less, and 14.9% never.

  • Multiple times a day: 9.2%
  • Once a day: 7.5%
  • 2–6 times a week: 12%
  • Once a week: 7.5%
  • 2–3 times a month: 11%
  • Once a month or less: 35.7%
  • Never: 14.9%
Share of respondents by outage frequency, 2026. Weekly or more (highlighted) totals 36.2%. Rows sum to 97.8%; the report does not explain the rest. Survey run with ETR, fieldwork April–May 2026. New Relic sells observability. Source: New Relic 2026 Observability Forecast, p.18.

Failures, detection time and cost

  • 82%

    Of respondents hit at least one production failure tied to AI-generated code in the past six months.

    New Relic × Hanover, State of AI Coding 2026 · 200 US respondents, vendor-funded

  • 41.3 min

    Mean time to detect, per the 2026 Forecast. Mean time to resolve averages 53.6 minutes.

    New Relic 2026 Observability Forecast · vendor-funded

  • 57%

    Of respondents say their most recent major outage cost over $100,000. 1 in 5 say over $1 million.

    Uptime Institute Annual Outage Analysis 2026 · independent

Rated well, then shipped

  1. Perceived

    Quality

    94% rate AI-generated code higher quality than human-authored code (61% somewhat, 33% much).

  2. Shipped

    Unchecked

    62% often ship AI code without checking it line by line.

  3. Production

    Incidents

    78% report more production incidents over the past 12 months.

  4. Production

    Senior time

    86% report more senior-staff time spent fixing code.

  5. Production

    Failures

    82% hit a production failure tied to AI code in six months.

Same 200 US respondents, all at companies using AI for development. The 78% and 86% are environment-wide trends; the survey does not attribute them to AI code. New Relic × Hanover, 2026, vendor-funded.

Our view: code that looks good in review can still break under real traffic, and only production shows it. The survey reports these numbers side by side. It does not show that one causes another.

What to do

  1. Consider a release unproven until it has run under real traffic. Ship behind flags and canaries.
  2. Alert on detection and resolution time as well as uptime.
  3. Make AI-generated code emit traces and logs by default, so you can see in production what review missed.

Questions

How often do companies have high-impact outages?
36% of 2,575 respondents to New Relic's 2026 Observability Forecast report a high-impact outage weekly or more; 14.9% report none.
How long do outages take to detect and resolve?
Respondents to New Relic's 2026 Observability Forecast report a mean time to detect of 41.3 minutes and a mean time to resolve of 53.6 minutes.
How much does a major outage cost?
57% of Uptime Institute respondents said their most recent major outage cost more than $100,000; 1 in 5 said over $1 million.

Sources