Use It Or Lose It?

A couple of weeks ago, I wrote an article about Vibe Coding, which gathered a modest amount of interest thanks to the salience of the topic and some welcome mentions by John Allspaw and others. One aspect in particular seemed to strike a chord with many readers: the concept of complementary and competitive cognitive artefacts.

Across technology, and incident response in particular, the key question is whether our tools augment or outsource human cognition. In the current AI gold rush, we risk trading understanding for speed - with notable consequences for resilience.

To spare you the repetition of reading yet another article on vibe coding, I’ll briefly summarise the concept of complementary and competitive cognitive artefacts here.

Complexity Scientist David Krakauer categorises tools into 2 types: Complementary Artefacts and Competitive Artefacts. The best way of describing the difference is by example:

A map is an example of a Complementary Cognitive Artefact. It is complementary because it not only serves as a tool to navigate from A to B, but it also enhances the user’s own mental model of the route and terrain upon which they’re travelling. The map is an accurate model of reality at a convenient level of fidelity that remains in the cognition of the diligent user after it’s no longer physically in view.

Satellite Navigation, on the other hand, is an example of a Competitive Cognitive Artefact. It is competitive because it although it does a (mostly) marvellous job of getting us from A to B with less effort required than the map reader, it does nothing to build our mental model of the route or terrain on which we’re travelling. If you’ve ever found yourself lost without sat-nav while navigating a route that you’ve previously done many times before with sat-nav, you get the picture.

Other commonly cited examples are: -

Calculator - Competitive; Abacus - Complementary

While a calculator enables rapid and accurate answers to mathematical calculations, it teaches us little about maths other than how to use a calculator. In contrast, an abacus facilitates mathematical calculation, but the physical manipulation and visualisation of beads helps the user to structure how to think about mathematics.

GarageBand Smart Piano - Competitive; A Piano - Complementary

Smart Instruments, such as those found in GarageBand, provide a convenient facsimile of the sound of a real instrument. It enables arpeggiation and stylistic chord progressions without surrendering any useful information about how such sonic wonders might be created outside of GarageBand.A piano, on the other hand, provides the capability to do all of the above, plus it offers a model with which to internalise the structure of music in its linear arrangement of pitches and spacing between pitches from left to right.

Complementary > Competitive?

It’s tempting to broadly appraise complementary cognitive artefacts as ‘good’ and competitive as ‘bad’, but this would be a vast oversimplification. Only a hardened Luddite would argue that a DJ requiring a piano sound needs to be a concert pianist. Equally few would argue that a concert pianist should spend less time tickling the ivories and more time thumbing GarageBand presets.

Rather, one’s choice of artefact depends on one’s needs, both now and in the future. In Joint Cognitive Systems (such as organisations), artefacts need to meet the needs of many people and will be constrained and directed by their intersection with technology.

This brings us to incident response and the artefacts that we rely on to keep things running, and to facilitate the efforts to fix things when they inevitably refuse to stick to the script.

Cognitive Artefacts in Incident Response

IT incidents are high-stakes and time-pressured; they require effective coordination among responders, associated stakeholders and the technology.

Such challenges rely on cognitive artefacts to inform diagnosis, interventions and to maintain common ground amongst responders. It’s rare for a single person to have all the necessary knowledge to handle an incident alone, and this requirement for distributed cognition is bolstered by the use of cognitive artefacts.

Such artefacts include: -

Logs / telemetry

• Metrics: numeric time-series (latency, error rate, throughput, resource utilisation, etc).
• Logs: event records, often text, with varying structure.
• Traces: distributed request flows showing causal paths across services.
• Events: discrete system-reported changes (deployments, alerts firing, config changes)..

Constructed artefacts

• Dashboards: curated visual groupings of metrics/traces for monitoring or exploration.
• Alerts: rules or anomaly detectors triggering notifications.
• Service level indicators (SLIs) and error budgets: derived performance artefacts tied to SLOs.

Collaborative artefacts

• Runbooks / playbooks: documented guides for interpreting and acting on signals.
• Incident timelines: constructed narratives of what happened when.
• Post incident reports: descriptions of previous incidents, causal factors and remediation
• ChatOps sessions/transcripts: conversational logs during operations.

Representational artefacts

• System / Architectural Diagrams: An architectural representation of the system upon which the response team is working.
• Org Charts: Who’s who and who does what.
• On Call Schedules: Who to contact, about what, and when.

Are these artefacts complementary, competitive or somewhere in between? I’ll wait…

Most of these are complementary artefacts, as they help practitioners to understand the current state of the system (at a useful level of fidelity), and provide a scaffold upon which the mental models of the practitioners can be built.

Competitive Cognitive Artefacts in Incident Response

As with Vibe Coding, AI serenades us with the siren song of faster resolution times, automated diagnostics and restoration. Some such approaches may be considered to be competitive cognitive artefacts, which, while being neither implicitly positive nor negative, are worth being aware of.

Such artefacts include: -

Automated root cause analysis tools (e.g., “probable cause: service X”)

• Replace diagnostic reasoning with automated causal analysis.

AI-based auto-remediation systems

• Solve issues automatically, perhaps without requiring responders to understand what failed or why.

Automated incident summaries

• Summarising ChatOps transcripts, timelines and interventions.

Auto-generated Post Incident Reviews

• Summarising learnings following an incident and suggesting follow-up actions to reduce the risk of recurrence.

Such tools sound alluring. Imagine how much more sleep we’d get if our AI assistants could be called at 3am instead of us. Imagine how much more productive we’d be if we didn’t have to attend long post-incident review sessions.

The future is nothing if not exciting, and none of us can say with certainty where this is heading, but here are some considerations…

It's not clear yet whether such tools will reduce responder cognitive load

AI-based causal analysis will suggest causes, and perhaps justification for these suggestions. However, practitioners will still need to evaluate whether these suggestions are worth listening to. The same issues concerning fixation and red herrings will occur regardless of whether a suggestion comes from a human or an AI. It remains to be seen whether fixation on AI suggestions, but research around AI’s impact on critical thinking hints that it might.

Also, until we’ve been entirely assimilated into the matrix, humans won’t be sitting idle while an AI handles the hypothesis generation; rather, they’ll be doing some thinking of their own:

The Left-Over Principle

Let’s say in 18 months' time, 90% of incidents in your company are diagnosed and resolved automatically. The remaining 10% left over will likely be the most complex (and therefore perhaps the most impactful). Given that practitioners have delegated 90% of their experience, expertise and context to AI, how might they be expected to effectively cope with the remaining wicked 10%?

Paradoxically, the requirement for human expertise increases as the proportion of automated ‘low hanging fruit’ increases until the moment that there’s nothing left over for the humans to do. Will that moment ever arrive?

There is Value in Artefact Creation as well as its Existence

“Life’s a journey, not a destination”. Cognitive artefacts don’t just surrender value due to their existence and usage. The process of their creation is also valuable in requiring practitioners to structure, organise and prioritise their tacit knowledge, to render it in an external form that makes sense to others.

A post-incident review or write-ups, for example, benefits the distributed cognition of the entire response team through their participation in its creation. There’s a magic in the grappling to make sense of an incident that can likely not be replicated if the finished article is presented without human effort. What you learn along the way is as valuable as the final destination, and AI can either help or hinder that.

Ultimately, the short and long-term utility of the new breed of AI-based tooling will be determined through experience, and early evidence from other domains is mixed. The lens of complementary and competitive cognitive artefacts is useful in grounding one’s thinking when surfing the hype wave.

One thing’s for sure: humans will remain indispensable in incident response for the foreseeable future, and they will need to find ways of practising alongside automated AI tooling to achieve the best results for themselves, their organisation, and their customers.

Uptime Labs helps organisations navigate this shift through incident response training that complements responder cognition, builds muscle memory, improves decision quality and communications under realistic pressure - the types of skills that you’ll need with, or without AI.

Further Reading

Hot off the presses: InfoQ has recently published a virtual panel: How Software Engineers and Team Leaders Can Excel with Artificial Intelligence. Courtney Nash, in particular, does an amazing job of breaking down the challenges and opportunities of working alongside AI and offers a long list of academic papers, lending a much-needed scientific lens to a topic that’s frequently overpowered by opinion.

*header image credit: Photo by @glenncarstenspeters on unsplash