I recently read Spec-Driven Development by Haim Michale, and it resonated with some day-to-day thoughts about SDD.
The problem SDD aims to solve isn’t new
Missing, incomplete, and ambiguous requirements that evolve over time have always plagued software delivery, but it is now amplified: the pace is faster, LLMs have a strong tendency to fill gaps by guessing, and they carry zero tribal knowledge – no kitchen-table conversations, no “oh, we tried that in 2022” instincts. EARS syntax (Easy Approach to Requirements Syntax) is an interesting framework for making requirements more machine-readable, but it still requires a human to steer it. The spec is only as good as the intent behind it.
Legacy code and brownfield projects are the elephant in the room
Most SDD content assumes you’re starting fresh. But the vast majority of engineering work happens in systems with years of accumulated decisions, undocumented assumptions, and implicit context baked into the codebase. Making that accessible to coding agents, let alone to SDD, is a real and largely unsolved problem. As far as I’m aware, there are no clear best practices or guidelines here yet. This is an area we need to tackle head-on.
SDD also accelerates a role blur we’re already talking about
If specs are first-class artifacts that live in the repo, who writes them? Do PMs start pushing requirements documents directly into codebases, expanding into territory traditionally owned by engineers? Do developers take on deeper responsibility for refining and maintaining specifications? Both are plausible. Neither is cost-free. The boundary is getting blurry fast.
And drift…
Drift is what happens when specs and code silently diverge: a requirement changes, but the spec file doesn’t, or a PR lands and the corresponding document isn’t updated. Over time, specs stop describing the system, and that’s arguably worse than having no specs at all, because they create false confidence. My suggestion: drift detection should be treated as a core part of repo gardening. Flag when code changes without a corresponding spec update, and vice versa. Make it visible, not something that accumulates quietly in the background. Expect spec debt terminology coming soon.
While different SDD methodologies vary in the exact documents they produce, I found the attached screenshot a useful cheat sheet for knowing when to update which document.
Coding agents are no longer a novelty – they’re everywhere. Over the past year, we’ve seen massive adoption across startups and enterprises, alongside real improvements in autonomy, reasoning depth, and multi-step code execution. Tools like Claude Code, Codex, Copilot, and Kiro are shipping updates at a relentless pace, and teams are increasingly comfortable letting agents refactor modules, write tests, and manage pull requests.
But there’s a catch: these tools are token eaters. Autonomous agents don’t just answer a prompt – they plan, reflect, re-read the codebase, call tools, retry, and iterate. At scale, that translates into serious API bills.
That’s why we’re seeing growing interest in a different deployment pattern: running coding agents against local or self-hosted models. Ollama recently announced ollama launch a command that sets up and runs coding tools such as Claude Code, OpenCode, and Codex with local or cloud models. vLLM, LiteLLM, and OpenRouter also provide similar integrations. That signals that this is no longer fringe experimentation. For many teams, local LLMs are emerging as a viable path to reduce cost, improve stability, and gain tighter control over privacy.
Deployment models for coding agents
When teams talk about “running models locally,” they often mean different things. In practice, there are three distinct deployment patterns – and they differ meaningfully in cost structure, performance profile, and governance posture.
Local (Developer Machine) – the model runs directly on a developer’s laptop or workstation (e.g., via Ollama).
Hosted (Org-Managed Infrastructure / VPC) – the organization runs the model on its own infrastructure, either on-premises GPU servers or in a private cloud/VPC (e.g., via vLLM, Kubernetes, or managed GPU clusters).
Managed LLM API (e.g., Anthropic, OpenAI, etc.) – the model runs fully managed by a provider; the organization interacts via API.
Dimension
Local (Dev Machine)
Hosted (Org VPC / On-Prem)
Managed LLM API
Cost Structure
No per-token fees. Hardware cost borne by the developer. Cheap at a small scale; uneven across the team.
No per-token fees. Significant infra + ops cost. Economical at scale if usage is high.
Usage-based (per token / per request). Predictable but can become very expensive with agent loops.
Cost at Scale (Agents)
Hard to standardize; limited by laptop GPU/CPU.
Strong cost efficiency at high volume
Token costs compound quickly. Expensive in large org rollouts.
Performance (Latency)
Very low latency locally, but limited by hardware. Large models may be slow or impossible.
Good latency if well-provisioned GPU cluster. Can optimize with batching.
Typically excellent latency and throughput; globally distributed infra.
Model Size / Capability
Limited to smaller models (7B–34B typically; maybe 70B with strong GPUs).
Can run large open models (70B+), depending on infra budget.
Access to frontier SOTA models (often strongest reasoning & coding quality).
Quality (Coding Tasks)
Improving. “Good enough” for many workflows, especially with fine-tuned coding models.
Strong – can choose best open models and fine-tune internally.
Often highest raw reasoning quality and reliability on complex multi-file tasks.
Security / Privacy
Code never leaves device. Strong for IP protection. Risk: inconsistent security posture across developers.
Lightweight CLI + API that serves models locally; integrates with multiple agents (Claude Code, Codex, Droid, OpenCode) and supports on-prem inferencing with moderate hardware.
vLLM (Serving)
High-performance LLM server
Optimized for scalable reasoning and long context LLM inference; integrates with agents (e.g., Claude Code) via Anthropic-Messages API compatibility.
OpenRouter
Unified LLM API broker
Central API layer for 400+ LLMs including local and cloud endpoints; can route agents to preferred backends with cost/redundancy optimization.
LiteLLM
Unified LLM API
Enables developers to use many LLM APIs, such as OpenAI, Anthropic, Gemini, and Ollama, in a single, OpenAI-compatible format.
Notable models
Model
Primary Use
Latest Release
Qwen3-Coder
Alibaba’s 480B-parameter MoE coding model. SOTA results among open models on agentic coding tasks
Cost is the most immediate driver. Autonomous coding agents are token-intensive by design. At enterprise scale, those token costs compound quickly.
Local inference eliminates per-token fees, which makes it attractive for high-volume, repetitive tasks. But frontier proprietary models still maintain an edge on complex, cross-repository reasoning and edge cases. The likely outcome is not full replacement, but intelligent routing:
Simpler or repetitive tasks → local or hosted open models
Tools like OpenRouter and LiteLLM are already enabling this pattern, and by the end of 2026, hybrid routing is likely to be the default deployment strategy for medium- to large-sized engineering organizations.
2. Standardization Will Lower the Switching Cost
Hybrid only works if switching models is frictionless.
As coding agents like Claude Code, Codex, Copilot, and others converge around shared inference interfaces (Ollama, vLLM, OpenAI-compatible endpoints), swapping models in and out becomes operationally simple. This reduces lock-in and makes experimentation safer. As interoperability improves, the barrier to trying local models drops dramatically – and adoption follows.
3. Open-Source Coding Models Will Close the Gap
Tool-use fine-tuning is maturing. Code reasoning benchmarks are becoming more rigorous.
By late 2026, open-weight coding models are likely to be “production-grade” for a substantial share of workflows – especially where cost control and data sovereignty matter more than absolute frontier performance.
4. Resilience Will Matter as Much as Cost
There’s also a structural pressure building: agent-driven workloads amplify the impact of API outages. When a coding agent is embedded into CI pipelines or developer workflows, downtime is no longer an inconvenience – it’s a blocker.
As usage scales, reliance on a single managed API becomes a risk vector. This will accelerate investment in redundancy:
Secondary API providers
Local fallback models
On-prem capacity for critical workflows
Summary
In 2026, hybrid won’t just be about cost optimization – it will be about operational resilience.
The future is not “local vs cloud.” It’s a composable, policy-driven model infrastructure.
Organizations that treat model routing, hosting strategy, and redundancy as part of their core engineering architecture – rather than as an afterthought – will have structural advantages in cost control, privacy, and reliability.
2026 won’t be the year enterprises abandon managed APIs. It will be the year they stop depending on them exclusively.
🚀 AI is moving deeper into digital health – over the past week, both OpenAI and Anthropic have introduced major features aimed at bringing powerful AI capabilities into healthcare and life sciences. Links in the first comment
🔹 OpenAI: ChatGPT Health OpenAI has launched ChatGPT Health, a dedicated health experience that lets users securely connect their medical records and wellness app data (e.g., Apple Health, Function, and MyFitnessPal) to get more informed insights about their health and wellness. The feature is designed to help people better interpret test results, prepare for doctor visits, and navigate everyday health questions — not replace clinicians. Enhanced privacy protections ensure that health chats and data remain isolated and encrypted, and that users retain full control over connections and data.
🔹 Anthropic: Claude for Healthcare & Life Sciences Following an earlier announcement regarding Claude for Life Sciences, Anthropic introduced Claude for Healthcare alongside expanded life science capabilities, bringing its Claude AI into regulated medical and scientific use cases. This includes HIPAA-ready infrastructure and connectors to industry data sources (like CMS coverage rules, ICD-10 codes, and NPI registries) to support tasks such as prior authorizations, claims management, and clinical documentation. Claude can also summarize medical histories and explain test results in plain language. On the life sciences side, new integrations with clinical trial, preprint, and bioinformatics platforms aim to accelerate research workflows and regulatory documentation.
Both announcements show the AI industry racing into digital health with different focus areas. OpenAI’s move toward personalized health guidance for individuals complements Anthropic’s broader, enterprise-oriented tools for providers and researchers. Together, they raise exciting possibilities and important questions about regulatory standards, data privacy, and the role of AI in care delivery.
Bonus – GrantFlow – a grant management platform that automates discovery, planning, and application workflows for researchers and institutions.
AWS re:Invent 2025 took place this week, and as always, dozens of announcements were unveiled. At the macro level, announcing Amazon EC2 Trn3 UltraServers for faster, lower-cost generative AI training can make a significant difference in the market, which is primarily biased towards Nvidia GPUs. At the micro-level, I chose four announcements that I find compelling and relevant for my day-to-day.
AWS Transform custom – AWS Transform enables organizations to automate the modernization of codebases at enterprise scale, including legacy frameworks, outdated runtimes, infrastructure-as-code, and even company-specific code patterns. The custom agent applies those transformation rules defined in documentation, natural language descriptions, or code samples consistently across the organization’s repositories.
Technical debt tends to accumulate quietly, damaging developer productivity and satisfaction. Transform custom wishes to “crush tech debt” and free up developers to focus on innovation instead. For organizations managing many microservices, legacy modules, or long-standing systems, this could dramatically reduce the maintenance burden and risk and increase employees’ satisfaction and retention over time.
AWS Lambda Durable Functions – Durable Functions enable building long-running, stateful, multi-step applications and workflows – directly within the serverless paradigm. Durable functions support a checkpoint-and-replay model: your code can pause (e.g., wait for external events or timeouts) and resume within 1 year without incurring idle compute costs during the pause.
Many real-world use cases, such as approval flows, background jobs, human-in-the-loop automation, and cross-service orchestration, require durable state, retries, and waiting. Previously, these often required dedicated infrastructure or complex orchestration logic. Durable Functions enable teams to build more robust and scalable workflows and reduce overhead.
AWS S3 Vectors (General Availability) – Amazon S3 Vectors was announced about 6 months ago and is now generally available. This adds native vector storage and querying capabilities to S3 buckets. That is, you can store embedding/vector data at scale, build vector indexes, and run similarity search via S3, without needing a separate vector database. The vectors can be enriched with metadata and integrated with other AWS services for retrieval-augmented generation (RAG) workflows. I think of it as “Athena” for embeddings.
This makes it much easier and cost-effective for teams to integrate AI/ML features – even if they don’t want to manage a dedicated vector DB and reduces the barrier to building AI-ready data backends.
Amazon SageMaker Serverless Customization – Fine-Tuning Models Without Infrastructure – AWS announced a new capability that accelerates model fine-tuning by eliminating the need for infrastructure management. Teams can upload a dataset and select a base model, and SageMaker handles the fine-tuning pipeline, scaling, and optimization automatically – all in a serverless, pay-per-use model. This customized model can also be deployed using Bedrock for Serverless inference. It is a game-changer, as serving a customized model was previously very expensive. This feature makes fine-tuning accessible to far more teams, especially those without dedicated ML engineers.
These are just a handful of the (many) announcements from re:Invent 2025, and they represent a small, opinionated slice of what AWS showcased. Collectively, they highlight a clear trend: Amazon is pushing hard into AI-driven infrastructure and developer automation – while challenging multiple categories of startups in the process.
While Trn3 UltraServers aim to chip away at NVIDIA’s dominance in AI training, the more immediate impact may come from the developer- and workflow-focused releases. Tools like Transform Custom, the new frontier agents, and Durable Functions promise to reduce engineering pain – if they can handle the real, messy complexity of enterprise systems. S3 Vectors and SageMaker Serverless Customization make it far easier to adopt vector search and fine-tuning without adding a new operational burden.
I recently came across two blog posts that made me think, especially in light of a sobering statistic I’ve seen floating around: a recent MIT study reports that 95% of enterprise generative AI pilots fail to deliver real business impact or move beyond demo mode.
One post is a conversation with Werner Vogels, Amazon’s long-time CTO, who shares lessons from decades of building and operating systems at internet scale. The other, from Docker, outlines nine rules for making AI proof-of-concepts that don’t die in demo land.
Despite their different starting points, I was surprised by how much the posts resonated with one another. Here’s a short review of where they align and where they differ.
Where They Agree
Solve real problems, not hype – Both warn against chasing the “cool demo.” Docker calls it “Solve Pain, Not Impress”, while Vogels is blunt: “Don’t build for hype.” This advice sounds obvious, but it’s easy to fall into the trap of chasing novelty. Whether you’re pitching to executives or building at AWS scale, both warn that if you’re not anchored in a real customer pain, the project is already off track.
Build with the end in mind – Neither believes in disposable prototypes. Docker advises to design for production from day zero—add observability, guardrails, testing, and think about scale early. Vogels echoes with “What you build, you run”, highlighting that engineers must take ownership of operations, security, and long-term maintainability. Both perspectives converge on the same principle: if you don’t build like it’s going to live in production, it probably never will.
Discipline over speed – Both posts emphasize discipline over blind speed. Docker urges teams to embed cost and risk awareness into PoCs, even tracking unit economics from day one. Vogels stresses that “cost isn’t boring—it’s survival” and frames decision-making around reversibility: move fast when you can reverse course, slow down when you can’t. Different wording, same idea: thoughtful choices early save pain later.
Where They Differ
Scope: the lab vs. the long haul – Docker’s post is tightly focused on how to build POCs in the messy realities of AI prototyping and how to avoid “demo theater” and make something that survives first contact with production. Vogels’ advice is broader, aimed at general engineering, technology leadership, infrastructure, decision-making at scale, and organization-level priorities. Vogels speaks from decades of running Amazon-scale systems, where the horizon is years, not weeks.
Tactics vs. culture – Docker’s advice is concrete and technical: use remocal workflows, benchmark early, add prompt testing to CI/CD. Vogels is less about specific tools and more about culture: engineers owning what they build, organizations learning to move fast on reversible decisions, and leaders setting clarity as a cultural value. Docker tells you what to do. Vogels tells you how to think.
Organizational Context and Scale – Docker speaks to teams fighting to get from zero to one—making PoCs credible beyond the demo stage. Vogels speaks from AWS’s point of view, where the challenge is running infrastructure that millions rely on. Docker’s post is about survival; Vogels is about resilience at scale.
What strikes me about these two perspectives is how perfectly they complement each other. Docker’s advice isn’t really about AI – it’s about escaping demo hell by building prototypes with production DNA from day one. Vogels tackles what happens when you actually succeed: keeping systems reliable when thousands depend on them. They’re describing the same journey from different ends. Set up your prototypes with the right foundations, and you dramatically increase the odds that your product will one day face the kinds of scale and resilience questions Vogels addresses.
Over the past few months, I have been constantly reading conversations about how Generative AI will reshape software engineering. On LinkedIn, Twitter, or in closed professional groups, engineers and product leaders debate how tools like Cursor, GitHub Copilot, or automated testing frameworks will impact the way software is built and teams are organized.
But the conversation goes beyond just engineering practices. If we zoom out, AI will not only transform the workflows of software teams but also the structure of companies and even the financial models on which they are built. This kind of change feels familiar – it echoes a deeper historical pattern in how science and technology evolve.
Kuhn’s Cycle of Scientific Revolutions
During my bachelor’s, I read Thomas Kuhn’s The Structure of Scientific Revolutions. Kuhn argued that science does not progress in a linear, step-by-step manner. Instead, it moves through cycles of stability and disruption. The Kuhn Cycle1, as reframed by later scholars, breaks this process into several stages:
Pre-science – A field without consensus; multiple competing ideas.
Normal Science – A dominant paradigm sets the rules of the game, guiding how problems are solved.
Model Drift – Anomalies accumulate, and cracks in the model appear.
Model Crisis – The old framework fails; confidence collapses.
Model Revolution – New models emerge, challenging the old order.
Paradigm Change – A new model wins acceptance and becomes the new normal.
The Kuhn Cycle Applied to Software Development
Normal Science
For decades, software engineering has operated under a shared set of practices and beliefs:
Clean Code & Best Practices – DRY, SOLID, Unit Testing, Peer Reviews.
Agile & Scrum – Iterative sprints and ceremonies as the “right” way to build products.
DevOps & CI/CD – Automation of builds, deployments, and testing.
Organizational Structure – Specialized roles (frontend, backend, QA, DevOps, PM) and a belief that more engineers equals more output.
The underlying assumption is hire more engineers + refine practices → better and quicker software.
Model Drift
Over time, cracks began to show.
The talent gap – demand for software far outstrips available developers.
Velocity mismatch – Agile rituals can’t keep pace with market demands.
Complexity overload – Microservices and massive codebases create systems that are too complex for a single person to comprehend fully.
These anomalies signaled that “hire more engineers and improve processes” was no longer a sustainable model.
Model Crisis
The strain became obvious:
Even tech giants with thousands of engineers struggle with code sprawl and coordination overhead.
Brooks’ Law bites – adding more people to a project often makes it slower.
Business pressure grows – leaders demand faster iteration, lower costs, and higher adaptability than human-only teams can deliver.
Early AI tools, such as GitHub Copilot and ChatGPT, reveal something provocative – machines can generate boilerplate, tests, and documentation in seconds – tasks once thought to be unavoidably human.
This is where many organizations sit today – patching the old paradigm with AI, but without a coherent new model.
Model Revolution
A new way of working begins to take shape. Here are some already visible in experimenting, we can all see around us –
AI-first engineering – using AI agents for scaffolding code, generating tests, or refactoring large systems. Humans act as curators, reviewers, and high-level designers.
Smaller, AI-augmented teams
New roles and workflows – QA shifts toward system-level validation; PMs focus less on ticket grooming and more on problem framing and prompting.
Org structures evolve – less siloing by specialization, more “AI-augmented full-stack builders.”
Economics shift – productivity is no longer headcount-driven but iteration-driven. Cost models change when iteration is nearly free.
Paradigm Change
In the coming years, some of the ideas above, and probably additional ideals, could stabilize as the “normal science” of software development and organizational building. But we are not yet there. Once we get there, today’s experiments will feel as obvious as Agile sprints or pull requests do now.
We are in the midst of model drift tipping into crisis, with glimpses of revolution already underway. Kuhn’s lesson is that revolutions are not just about better tools – they’re about shifts in worldview. For AI, the shift might be that companies will no longer be limited by headcount and manual processes but by their ability to ask the right questions, frame the correct problems, and adapt their models of value creation.
We are moving toward a future where the shape of companies, not just their software stacks, will look radically different, and that’s an exciting era to be a part of.
Have you noticed how half the posts on LinkedIn these days feel like they were written by an LLM – too many words for too little substance? Or how product roadmaps suddenly include “AI features” that nobody asked for, just because it sounds good in a pitch deck? Or those meetings where someone suggests “let’s use GPT for this,” when a simple SQL query, an if-statement, or a much simpler ML model would do the job?
Laurence Tratt recently coined the term LLM inflation to describe how humans use LLMs to expand simple ideas into verbose prose, only for others to shrink them back down. That concept got me thinking about a related phenomenon: LLM debt.
LLM debt is the growing cost of misusing LLMs — by adding them where they don’t belong and neglecting them where they could help.
We’re all familiar with technical debt, product debt, and design debt1 — the shortcuts or missed opportunities that slow us down over time. Similarly, organizations are quietly accumulating LLM debt.
So, what does LLM debt look like in practice? It’s a double-edged liability:
Overuse: Integrating LLMs where they’re unnecessary adds latency, complexity, cost, and stochasticity to systems that could be simpler, faster, and more reliable without them. For example, sending every API request through a multimillion-parameter model when a simple regex or deterministic logic would suffice.
Underuse: Failing to adopt LLM-based tools where they could genuinely help results in wasted effort and missed opportunities. Think of teams manually triaging support tickets, writing repetitive documentation, or analyzing text data by hand when an LLM could automate much of the work.
Like product or technical debt, a small amount of LLM debt can be strategic: it allows experimentation, faster prototyping, or proof-of-concept development. However, left unmanaged, it compounds, creating systems that are over-engineered in some areas and under-leveraged in others, which slows product evolution and innovation. Same as other types of debt, it should be owned and managed.
LLMs are powerful, but they come with costs. Just as we track and manage technical debt, we need to recognize, measure, and pay down our LLM debt. That means asking tough questions before adding LLMs to the stack, and also being bold enough to leverage them where they could provide real value.
If LLM inflation showed us how words can expand and collapse in unhelpful cycles, LLM debt shows us how our systems can quietly accumulate inefficiencies that slow us down. Recognizing it early is the key to keeping our products lean, intelligent, and future-ready.
I previously wrote about some of those topics here and talked about them in Hebrew here↩︎
Agents app design pattern – this is a back-to-basics adaptation. How would we read this 14 years from now? Would the ideas he mentioned there be a standard?
When the Agents Go Marching In: Five Design Paradigms Reshaping Our Digital Future
This post complements the previous one, covering the same topics. If you are in a hurry, jump to the “The Reinvention of UX: Five Emerging Paradigms” section. I feel that I cope with all those aspects, e.g., building trust, transparency, cognitive load distribution, etc., on a daily basis.
Using Alembic to create and seed a database Seeding a database is essential for testing, development, and ensuring consistent application behavior across different environments. Alembic is a lightweight database migration tool for Python, designed to work seamlessly with SQLAlchemy.
We use Alembic to manage our database migrations, and I recently needed to seed our database for consistency across environments. I looked for several solutions and eventually used the solution in this post to create a migration that seeds the database –
A Field Guide to Rapidly Improving AI Products – while this post focuses on AI products, specifically ones LLM-based, multiple lessons can also be adapted to non-LLM-based AI products and general products. Conducting an error analysis, generating synthetic data (preferably with domain express), and using a data viewer are good starting points.
I Tried Running an MCP Server on AWS Lambda… Here’s What Happened– this post involves two topics I think a lot about these days – MCP and serverless computing. I think it is clear why I think a lot about MCPs. But why do I think about serverless computing? I think of it as a low-cost solution for early-stage startups. Early-stage startups usually have low traffic, which does not justify the cost of having servers up 24/7. On the other hand, the serverless development experience still needs some refinement, and there are services that one would like to host that do not support running in a serverless manner.
Tom Ron 