Eric J. Smith

Founder · Distinguished Engineer · Builder

Decades of judgement,
amplified by AI.

I've been writing software since 1991 and I'm still a builder at heart. Along the way I bootstrapped two software companies, and today I use AI agents to take on projects that once needed a whole team. The part that hasn't changed is knowing what's worth building.

Get in touch See my work
  • 2 bootstrapped companies
  • $16M+ product revenue
  • 87M+ package downloads
  • 6,000+ GitHub stars
now
Distinguished Engineer at Concord Finance
founded
Exceptionless & CodeSmith Tools
focus
Adaptable platforms and AI-amplified engineering teams

Today

Replacing a 40-year-old system with a platform.

Distinguished Engineer · Concord Finance · 2023 – now

Concord provides credit servicing and software for specialty finance, with more than $60 billion in assets under administration. I led the greenfield build of Concord Link, the platform replacing a green-screen system that had run the business for more than four decades.

We started with almost no written business requirements, so the job was as much about understanding the business as writing code: learning how servicing really worked from the people doing it, and designing for needs nobody could spell out yet. Rather than build for one lending vertical, we built a platform. Like Salesforce for loan servicing, it adapts to any kind of loan, lease or line of credit, so the business can grow into new markets without starting over.

Concord Link is already live with clients, and in Concord's own words it's built and tested "with AI at every stage of development."

About Concord Link
  • 1,500+code reviews of teammates' work, setting the bar for a team of 20+ engineers
  • AI reviewEvery change is checked by five specialist AI reviewers (correctness, security, architecture, performance and code hygiene) before a person signs off. 1,000+ pull requests so far.
  • Secure AIStaff can ask AI assistants about live loan data, limited to exactly what they're already allowed to see

The two systems that made it adaptable

scripting & durable workflows

Business rules as configuration, not code

In loan servicing, every client's contract is a little different. I designed and built Concord Link's scripting engine on Jint so those differences live in configuration: sandboxed JavaScript rules for payment allocation, fees, interest, statements and collections. On top of it I built durable workflows that can wait days for a payment or an account event and then pick up exactly where they left off, with a visual builder so operations staff can design them without writing code.

  • Triggered by account events, schedules or staff actions
  • Durable workflows with suspend, resume and deterministic replay
  • Scenario tests, versioning and AI-assisted authoring
  • Workflow core released as open-source Jint.Workflows

imports & exports

Meet clients where they are

Servicing runs on file formats that haven't changed in decades, and clients weren't going to change them for us. So I designed and built an import and export pipeline that adapts to them: declarative field mappings plus scripted transforms, encrypted files over SFTP or S3, on a schedule. Onboarding a new client and migrating them off the legacy system became setup work instead of new engineering.

  • CSV, Excel, XML and JSON, with PGP encryption
  • Mappings built from a sample file, with live preview
  • Record-level errors, retries, de-duplication and audit trail
  • The repeatable migration path off the legacy system

Earlier

  • Chief Software Architect, SlideRoom / Liaison International: led a greenfield admissions platform used by hundreds of colleges
  • Senior Database Engineer, Match.com
  • Architect & Lead Engineer, RedRiver Systems
  • Architect, Rare Medium
  • Software Developer, CurrSoft & Frontier Communications (from 1991)

Read the full résumé

The founder's lens

Twenty years of running the business, not just writing the code.

I bootstrapped two software companies, CodeSmith Tools and Exceptionless, with no outside capital. Over two decades their products brought in more than $16 million, and I was responsible for all of it: the code, pricing, sales, support, servers and payroll. Every technical decision eventually shows up in a support ticket, a hosting bill or a renewal.

Founder & President · 2003 – now

CodeSmith Tools

From a free side project to a two-decade business.

CodeSmith Generator started as a free tool in the earliest days of .NET and grew into one of its first widely used code generators. I built the template language, parser, compiler and IDE, and ran the business around it.

  1. 2003Free release; adopted across the early .NET community
  2. 2004Featured in MSDN Magazine's "Ten Must-Have .NET Tools," first paid edition
  3. 2006–09Visual Studio Magazine Readers' Choice and DevProConnections Best Utility awards
Years
20+
Awards
6
Funding
None
codesmithtools.com

Founder · 2013 – now

Exceptionless

An open-source project and a SaaS business at the same time.

Real-time error, log and feature-usage monitoring, hosted or self-hosted. Running it means balancing an open-source community with paying customers, and high event volumes with the hosting bill.

  1. 2013Founded and launched the hosted service
  2. 2014Open-sourced the entire platform on GitHub
  3. 2026Complete UI rewrite, plus AI that can investigate errors for you
NuGet
34M+
Docker
3.5M+
Stars
3.5k
exceptionless.com

Reading between the lines

A feature request is rarely about the feature. Behind "can you add an export button?" there might be a customer who doesn't trust the numbers, a sales deal that's stuck, or a workflow nobody understood.

Running businesses taught me to look at a tiny technical detail, step all the way back to see how it fits the product, the customer and the business, and then come back with the question behind the question. It's the skill I lean on most at Concord.

  1. BusinessIs it worth what it will cost to build, support, sell and run for years?
  2. CustomerIs it what they actually need, or just what they asked for?
  3. ProductDoes it fit with everything else we've built?
  4. FeatureDoes it solve the problem as described?
  5. CodeIs it correct?
AI has made the bottom rungs much faster to climb. The value has moved up the ladder, and that's where judgement matters most.

AI, judgement and the build loop

AI is an amplifier. It doesn't care what it amplifies.

Give an experienced engineer with good instincts an AI agent and you get something close to a small team. Give the same agent to someone who doesn't understand the business, the customer or the real cost of complexity, and you get the wrong thing, built faster and explained more confidently. Without a grasp of good design, they'll also dig a tech-debt hole so deep that even AI will struggle to climb out of it.

That's why architecture and judgement matter more than ever, and why I believe three skills deserve equal weight when we hire, promote and mentor engineers:

Business sense
Weighing what something is worth against what it will cost to build, run and support.
Common sense
Noticing when something doesn't add up, even when the AI sounds confident.
Technical depth
Designing systems that stay easy to change, and telling good output from merely plausible output.
AI widens the gap between builders How I see it · value created over time

Without AI, the difference between a strong hire and a poor one adds up slowly. With AI, both move many times faster: one compounds value, the other compounds debt, wrong features and risk. That's why hiring for judgement, business sense and common sense, not just technical skill, matters more than ever.

How I put it to work

Anyone can get an AI agent to produce code. Getting the right thing, reliably and at scale, takes a system. I build complete loops where agents plan, build, test, review and fix their own work against automated checks, so we can take on far bigger projects and ship far more without letting quality or consistency slip.

  1. 01SpecClear outcome, guidelines and skills the agents follow
  2. 02BuildSeveral AI agents working in parallel, each on its own copy of the code
  3. 03VerifyAutomated builds, tests, quality scoring and UI checks
  4. 04ReviewSpecialist AI reviewers, then a human who knows the business
  5. 05FixAgents address every finding and run the loop again
  6. →ShipAutomated pipeline, preview environments, production
Verify → Review → Fix repeats until every check passes

Guardrails: AI gets no more access than the person using it, every change is traceable, and architecture and trade-offs stay with people. "The AI wrote it" is never an answer.

Daily toolkit Claude Code · Codex · T3 Code · MCP · .NET Aspire · Playwright

Open source & projects

Building in the open for over a decade.

Alongside Exceptionless, I maintain libraries that run in production at my companies and plenty of others.

Selected projects

All on GitHub
Foundatio 2.5k stars · 53M+ downloads Pluggable building blocks for distributed apps. Caching, queues, locks, messaging, file storage and jobs behind one set of abstractions, with Redis, AWS, Azure and RabbitMQ implementations. C# · .NET Foundatio.Mediator 25K+ downloads A convention-based mediator built on source generators and interceptors. No runtime reflection, near direct-call performance, and generated API endpoints. C# · Source generators Jint.Workflows In production Durable JavaScript workflows for .NET. Suspend, serialize and resume long-running orchestration with deterministic replay. Powers the workflow engine in Concord Link. C# · JavaScript

Also contributed to OpenTelemetry .NET (Elasticsearch and Redis instrumentation), Jint, ASP.NET Core and Svelte ecosystem libraries.

Eric J. Smith

Get in touch

Say hello.

Whether it's something you're building, a question about my work or an idea you can't stop thinking about, send me an email. I'd love to hear about it.

Email me LinkedIn GitHub X