The catastrophic creation from confluent conniptions
Question? How do you rouse a 220 BC Qin dynasty distinguished dignitary to deepest bureaucratic satisfaction?
Answer: Create a document unhindered by what it governs. Unblemished by the artifice of capricious existence, requiring no reviews, reconciliations, or reckonings with reality.
A document that exists purely as an administrative artefact of and for itself. Unfortunately, we are three cloud consultants in the first quarter of the 21st century. Dealing in endless updates of morphing designs, constant reconciliation of evolving requirements mixed with a willowing cycle of architectural approvals and appeals.
All for a high-level design to uplift a customer's AWS Landing Zone. Six months on a single Confluence page. No line of Terraform or CloudFormation written. In the early weeks, you could read it within a seven-minute Bugs Bunny cartoon short. By the end, it rivaled the length of The Fellowship of the Ring. Director's Cut. Four-hour edition.
We finished with grim faces, broken motivation and frayed emotions. None of us derived any satisfaction from the traumatic ordeal. Refusing to ever bear this baffling bureaucratic burden again, I was ready to give AI a go.
Chatbots: Erroneous experiment in erratic electronic exchanges
First experiment began with a simple test. Chit-chatting to a chipset chum a.k.a chatbot. "What standard backup tiers should be used in a standard landing zone, following Well-Architected Framework best practices?"
It was like summoning a spirit from 220 BC who’d been method-acting Liu Ling’s "In Praise of the Virtue of Wine" a bit too intensely. The answer was gloriously amusing and misguided: "Utilise three backup tiers: online, offline, and deep archive. Since deep archive is the cheapest, all backups should exclusively use this tier. You'll be a FinOps master."
I could already imagine the nightmare. A 3 AM emergency database restore for a high-traffic e-commerce platform. The crushing realisation that no online or offline backups exist. The excruciating 12-hour wait for a deep archive retrieval.
Everyone, from the grim faced engineer, the furious manager, and the emotionally frayed CTO. United in their cursing of the three cloud consultants who built backups on flawed AI advice.
Spec driven development: Tedious trial of tiresome typing
It was the season of specs. A system of sanctioned specifications, systematic planning, and staged builds. We were like Spring and Autumn period imperial bureaucrats. Elucidating Iron Age edicts, utterly defeated to ineffectiveness by Bronze Age knowledge. Suffering from a latency of wisdom.
Let me explain. Foundation models are trained in frozen history. Ask for an optimal EC2 class today and it'll return with a class from yesterday. It was an augur of ancient wisdom. Brilliant at predicting patterns of yore, but utterly blind to omens in the contemporary.
RAG(Retrieval-Augmented Generation) arrived to address this flaw. However, we found ourselves exporting reference web pages and Confluence pages into markdown. Then curating, collecting, and uploading the files. Instead of saving time, we had bartered one bureaucratic burden for another.
Agents, skills and MCP. Promising pioneering pathway
Agents, skills, and MCP(Model Context Protocol) allowed us to become the distinguished dignitary. Backed by a ministry of expert civil servants. Transmuting imperial edicts into cloud reality. No more chit-chatting with chipset chums announcing archaic arcana.
Gone, the systematic scribbling sustaining specifications scratchings. With an Architect Agent, I delegated an entire AWS Landing Zone design. Connected to AWS MCP, it pulled the latest Well-Architected Framework patterns. While custom skills defined the JIRA backlog and published the final design to Confluence.
Without bureaucratic design friction, skills also enabled building and auditing of Landing Zones. A custom skill could generate and deploy custom Landing Zone Accelerator code. While an AWS DevOps Agent could audit the state of an existing landing zone.
Conclusion
In a single whirlwind year, we evolved from the soul crushing trauma of endless Confluence page updates to redefining how Landing Zones are designed and built.
What began as a personal pursuit to preserve perspicacity paved a pathway to pioneering possibilities.
Stop writing Fellowship of the Ring length design docs. Become the distinguished dignitary. Don the imperial robes, and let your silicon department bear the bureaucratic weight.
Give AI a go in building and deploying an AWS Landing Zone. Start small. Design a backup solution. Generate the code. See the results for yourself, and see what's possible for your satisfaction.