Murali Suriar

Dave Anderson at Amazon: improve platform error rates without authority over execs Problem: he had data but no authority. The move: publish error rate by team, and owning VP. What happened: people rushed to him to get their team off the list. The measurement doesn't have to mandate change, it just makes status quo uncomfortable.

@garius hoooly shit. it's literally "all of this has happened before"

Have you been paged for something your tooling should have caught? How many have a runbook you don't trust? How many have had to context switch out of an incident to understand how tools/logging work? [many hands] you just measured your developer/SRE experience!

SREs are developers, with the same failure modes

• Write code
• Build tooling
• Manage complex multi-system workflows

Failure modes:

• cognitive overloap
• tool failures
• process bottlenecks

When SRE systems fail, the blast radius is often every incident the SRE is on

Different types of friction:

• Cognitive load: alerts without context; opaque system states; mental model gaps
• Tool friction: UI that fights us under pressure ; CLIs with undiscoverable flags ; dashboards that load slowly
• Process friction: coordination lag during incidents ; approval gates mid incident ; urn books that require interpretation

Friction doesn't just slow you down - it increases error rate under pressure. Jevons paradox turns friction up to 11. More complex systems means that cognitive, tool and process friction also increase, and increase much more quickly.

AI is shipping more code

• deployment frequency up
• incident rate follows
• more oncall pressure (more incidents and more friction)

[missed some slides]

SPACE for SRE:

• Satisfaction: oncall burden, tools
• Performance: MTTx, alert precision, accuracy under pressure
• Activity: toil ratio, automation percentage
• Comms: handoff quality, incident coordination overhead
• Efficiency: tool context-switching, flow interruption, time to understand...

Your team's experience is a leading indicator of your reliability outcomes

With AI:

• Efficiency goes up, but trust goes down (because your intuition is worse)
• Satisfaction: can I trust the robot?
• Communication: harder to hand off AI-generated context
• Performance: new failure modes.

New northstar metric: Mean Time to WTF Time to alert to "I understand what is happening" [ed: or rather "I have found the first thing that is obviously broken"]

Three low cost starting points:

• incidents retrospective tagging. friction contribution field!
• oncall survey, 2-3 questions. what slow down, what helped, why did you swear, what did you have to invent...
• toil tracking sprint. time box one sprint for fixing toil. Ask everyone to log manual, repetitive work (not fix, just log) [ed: ideally log this all the time?]

More takeaways:

• start a friction log
• time your MTWTF - once
• Declare a toil hour - log all the friction

Real costs of friction:

• $1.52T in tech debt, annually (CISQ)
• 40% of dev budgets of dev budgets are spent on avoidable rework (McKinsey)
• $300B in lost productivity (assuming devs self reported 68% productivity feeling is accurate)

But you need to reframe it.

Productivity --> Experience Productivity is about measuring output. Experience is about removing obstacles. For leadership: stop caring about devs working "hard enough" and start thinking about "removing what's slowing us down".

The trap: developer experience is not important. Reliability is the frame that leadership will fund. Friction --> Slower MTTD/MTTR --> Higher change failure rate --> SLA risk --> Revenue impact "We are proposing a better system - SRE satisfaction is a side effect". Anchor it to what leadership own. Don't build your cost model from scratch. Steal the one your finance team uses for downtime cost estimates.

Strategies for making the business case:

• Visibility and accountability: use competitive dynamics and public comparison
• Simple data, clear action: progressive disclosure, right detail at right time
• Dollar impact: speak the language stakeholders understand

Friction reduction list:

• open a runbook, see if it's real
• talk to a peer about what makes them swear
• start a friction log, look for patterns
• time your MTWTF on one real incident
• declare a toil error - inventory first, then fix
• ask your finance team what your downtime cost number is

What's next?

• IC: map your workflow for one week. Where do you lose 30+ minutes?
• Team leads: 30 minute retro. What slowed us down?
• Leaders: listen to people. coffee, whatever. Napkin math the costs of friction you hear about

more resources: https://bit.ly/frictionless-book?r=qr