Post by Polansoft

Working with AI is much like playing a sophisticated guessing game: for each input the models try to guess the best response. Given the nondeterministic nature of the AI, it is impossible to predict the outcome each time, but given the right infrastructure one can guide the AI from guessing and deriving to concrete discovery. One common pitfall I noticed is that we expect AI to produce the result on par with our own abilities, without access to knowledge base that we have. This is often attributed to either lack of capabilities of a model and being solved by a complex multi-layer agentic setup - which are very difficult to build and maintain on the early stages of the AI adoption. Instead, I suggest beginning by adopting the existing knowledge to the AI tech you use. Each team has an established set of rules, protocols and procedures, which drive the expectations: some live in confluence pages and formalized documents while other are built-in muscle memory that is passed to new hires with tech-review comments repeated in couple of pull requests. By formalizing existing knowledge into the accessible infrastructure, the team provides the access to the same baseline that any team member would have. This allows to share the common knowledge bank across the entire team, bake prompts for repeated taks to reduce the burden of prompt-generation on the users. The bottom line is — start from collecting and structuring the existing knowledge, not inventing new processes. Insight from Efosa Guobadia #Mainframe #MainframeModernization #LegacySystems #EnterpriseIT #DigitalTransformation #SoftwareEngineering #AICode #GenerativeAI #Innovation #TechLeadership #Copilot