“The hardest thing about getting started, is getting started.”
— Guy Kawasaki

‍

Adults make about 33,000 to 35,000 decisions a day. While 95 percent of those are made subconsciously, the remaining 5 percent percolate into our cerebral cortex and require System 2 thinking. System 2 thinking takes effort: it weighs pros and cons, risk and reward.

‍

Take a big decision, say, how to modernize a particular legacy system (naturally). Do we add to an already groaning legacy estate or do we bite the bullet and rewrite the application? Do we risk disrupting business as usual or do we propose a multi-year migration roadmap? Do we risk a 70 percent failure rate or keep duct-taping systems until it becomes someone else’s problem?

‍

Color us unsurprised when we found that 71 percent of technology leaders in research we recently commissioned agreed that, “the prospect of legacy modernization induces stress to IT leaders or yourself within your organization,” even as 74 percent saw legacy modernization as key to embracing innovation. (Subscribers get a first look at the findings next month!)

‍

In other words, it’s hard to get started when the confidence level for success is empirically low. Yet, there’s a simple way to build genuine, evidence-based confidence:

1. Focus on individual workflows. Also called complexity partitioning.
2. Select a single workflow, rewrite, and get it into production.
3. Continually, radically reduce the time subsequent workflows take to get into production.

‍

That’s it. Even though we can’t say for certain how much it will cost by the end of the entire migration, we do know what we’ve spent to get one workflow of n workflows into production because we’ve done it. It’s in production, it’s working. We then have a much higher confidence in projecting the cost of and time required for subsequent workflows.

‍

This trades long-term, often inaccurate, budget forecasts for precise spend data in short-term increments: an accelerating accretion of small successes actually in production that eliminates uncertainty.

‍

And that makes decisions safer — and less stressful.

‍

News and Views

A study commissioned by CIO.com has revealed that 71 percent of IT leaders are diving into GenAI — and 80 percent expect that their AI and ML projects will only increase over the next year. Too bad only a fraction of the data they need is accessible because... legacy.

‍

Leda Glyptis vividly brings to life just how hard it is to get going on big projects (ahem, AI) in her latest blog on Fintech Futures. (Also: she was right about mainframes on Mars.)

‍

Move over FOMO, say hello to FOMU — the fear of messing up, explained in Juniper Network's latest study. Despite organizations’ eagerness about AI, over 80 percent of those leading the charge feel rushed and very conscious of the mounting pressure to deliver.

‍

Modernizing a mainframe is best done holistically, as Legal & General has demonstrated. Jen Riggins from The New Stack explores their empathy-driven design thinking approach, steering them away from the classic “Big Bang” method and towards understanding the different groups working with the mainframe and their pain points.

‍

Retrieval-augmented generation (RAG) is a technique that improves the accuracy and reliability of generative AI models by retrieving facts from external sources. It provides a layer of "clean prompts" and fine-tuning. The Pragmatic Engineer showcases how engineering teams are increasingly building their own RAG pipelines for their LLMs.

‍

From the Orchard

Jeff Schomay previously wrote on how we use RAG to make AI more reliable. His latest blog expands on enhancing language model applications through structured outputs and validation for more efficient, reliable AI with minimal coding.

‍

Managing the risks inherent in legacy system modernization is crucial to success. Our CCO Edward Hieatt gives a piece of advice that’s both fresh and conventional: adopt a safety-first approach.

San Francisco is alive and well, and we’re doing our part. San Francisco Business Times covers our recent move to new digs.

‍

Curious to learn more? Say hello@mechanical-orchard.com.

‍

——————

Issue first published on May 30th, 2024.

‍