In this week’s blog, we look into the economics of causal reasoning engines, understand why self-driving cars have had difficulty commercializing, and why developer tooling is on the precipice of seeing breakthroughs.

The costs of any causal reasoning engine can be boiled down to three key components - a good user interface to understand the user’s request and relay back results, data collection units to understand the state of the world, and computing power to process said data based on the user’s requests. There are three kinds of economic benefits a business or an individual could expect by deploying an effective causal reasoning engine - boosting revenue by improving productivity, reducing expenses by replacing expensive tooling, and reducing opportunity cost by allowing employees to focus on higher-value tasks. For such technology to be economically viable, the benefits must far outweigh the costs since the development and sales cycles often go into years if not decades.

Self-driving cars were shuttling between San Francisco and Lake Tahoe as far back as 2011. However, they ran into several economic issues when dealing with the long tail of edge cases. First of all was the decision to use Lidars for depth perception which was a deal breaker for many of the edge cases involving fog and rain. However, concerns around cost forced many companies including Tesla to move forward with cheaper stereo cameras that provide lower fidelity. The need to run ML algorithms onboard the vehicle to satisfy data privacy and latency concerns while dealing with a global chip shortage didn't help matters either. This had deep geo-political concerns including the ability to produce chips domestically and countries are now scrambling to set up local chip manufacturing plants in case of disruption of chip manufacturing in Taiwan. Lastly, deciding the right level of user intervention has always been the albatross around the neck for self-driving vehicles. On one hand, they promise to provide drivers with the ability to free themselves from driving. On the other hand, they require drivers to always be ready to take over control at a moment's notice. The three concerns when taken into consideration with the fact that most Americans and Canadians already spend a significant amount of their disposable income on their vehicles, any additional cost to the sticker price means a reduction in the market size for such technology.

So do the benefits outweigh the costs for self-driving cars? The struggles with commercialization indicate that the answer is a negative. But all is not lost, we can take the learnings from the failure of self-driving vehicles. In my opinion, the domains where causal reasoning engines will be successful will be ones where:

• Data collection is primarily via scalable software integrations instead of expensive hardware sensors
• Compute resources are plentiful either in the form of multi-tenant cloud-based GPUs or allow using existing hardware in the enterprise
• User onboarding and off-boarding are clear, methodical, and done at a slow pace (measured in minutes, not milliseconds)
• Users are not price-conscious and/or have not already stretched their budget

Now let us see how the domain of developer tooling, especially incident investigation, fulfills these four criteria. On the data collection front, we have observability data freely available from almost every system across the stack at a great resolution. Second, most enterprises have large cloud budgets making it possible to run memory and compute-intensive workloads. Third, incident investigation is a methodical process that supports slow-paced user onboarding and offboarding. Lastly, open-source observability platforms such as Prometheus, Loki, and Open-telemetry make it possible for even large organizations to completely skip pricey, proprietary solutions such as DataDog or Splunk. This means that there is plenty of room for commercializing causal reasoning engines in this domain and I strongly believe that we are on the precipice of a breakthrough in developer tooling, especially incident investigation.

Stay tuned to learn more!

Deepak

Co-founder, Hoistr