Categories
Conferences Meetups Tampa Bay

Tampa Bay Tech Week: April 7 – 12!

Tampa Bay Tech Week is only a couple of weeks away!

Billed as “a multi-day, citywide celebration of technology, culture, and community,” it’s a week worth of events taking place across Tampa Bay, from Ybor and downtown Tampa to midtown Tampa to St. Pete.

The organizers have teams up with startups, enterprises, investors, and community organizations to put this series of events together, and it’s for techies of all stripes: founders, engineers, creators, and students.

Tampa Bay Tech Week’s events

Here’s the latest list of events associated with Tampa Bay Tech Week. They might add more, so to be sure, check the Tampa Bay Tech Week events page often!

Tuesday, April 7

Wednesday, April 8

Thurssday, April 9

Friday, April 10

Where to get passes

You can get passes from the Passes page on the Tampa Bay Tech Week site. There are different passes:

  • The free EXPO Pass, which gives you access to the Talent & Tech Expo.
  • The GA Pass ($150), which gives you access to all Tampa Bay Tech Week events, seating at Hotel Haya, entry to Interactive Workshop Panels™, discounted tickets to BŪP Innovation Weekend, and entry to community events across the region.
  • The VIP Pass ($300), which gets you additional access to BŪP Innovation Weekend, the VIP After Party, the Cyber + Cigars Networking Event, Closing Ceremonies, and VIP Seating at major sessions at Embarc Collective and Hotel Haya.
  • The All Access Pass ($450), which gets you exclusive access to the Tampa Bay Tech Week Yacht Event.

The Tampa Bay Tech Week Team

Tampa Bay Tech Week is organized by this team:


For more info about Tampa Bay Tech Week, visit the Tampa Bay Tech Week site!

Categories
Artificial Intelligence Humor

My conversations with AIs (#2 in a series)

Pictured above: a screenshot from a conversation I had with Claude last night. I was using it to craft an email in response to an interview that could be better described as an ambush.

I use LLMs as a double-check for when I’m trying to remain professional when greatly annoyed and for when I want to spend the minimum amount of time on something or someone. The party I was communicating with met both criteria.

I’m pretty sure that the way I sass back at LLMs doesn’t affect the quality of their revision when I harshly tell them that I disagree with their original answer. But as a catharsis, creative outlet, and excuse for an amusing screenshot and blog post, it’s oh-so-good.


Want more? Here’s a post from January about what I call The “Nick Fury” method for refusing LLM answers.

Categories
Artificial Intelligence Conferences Programming What I’m Up To

My favorite talk title from the upcoming Arc of AI conference (April 13 – 16)

From April 13th through 16th — and a couple of days before, because it’s in Austin — I’m going to be at the Arc of AI conference! Over the next little while, I’m going to be posting articles about Arc of AI, in case you’re wondering what the conference is about and whether you should go.

In this article, I’ll talk about my favorite title from all the talks on the Arc of AI agenda.

The talk: We’re all using AI, But We’re Not Enjoying It

When your talk happens on the last time slot at the end of a three-day conference (four days, if you’re also going to do one of the workshops), you need to put in some extra effort to get the attendees to show up and not disappear for the local sights (Arc of AI’s in Austin) or make a beeline for the airport.

Brent Laster, President and Founder of Tech Skills Transformations, is giving a number of talks — and a workshop! — at Arc of AI, and he has one of the closing talks. He has a talk in one of those last speaking slots on the Thursday at 4:00 p.m., and it has what I think is the most interesting title on the agenda:

We’re all using AI, But We’re Not Enjoying It

Here’s the abstract:

We’re All Using AI, But We’re Not Enjoying It takes an honest look at a growing gap in the workplace: AI adoption is skyrocketing, yet frustration, confusion, and uneven results are just as common. This talk explores why AI so often feels harder than it should—poorly integrated tools, unclear workflows, unrealistic expectations, cognitive overload, and the pressure to “keep up.” Looking at patterns seen across teams learning to use AI effectively, we’ll break down the practical barriers that make everyday AI work feel tedious instead of empowering. More importantly, we’ll outline a set of achievable shifts—better task design, lighter mental models, context-first prompting, workflow pairing, and small but meaningful guardrails—that can restore a sense of control and clarity.

I need to figure out how I can attend both Brent’s talk and my former Tucows coworker Leonid Igolnik’s talk (which he’s giving with Baruch Sadogursky), Back to the Future of Software: How to Survive the AI Apocalypse with Tests, Prompts, and Specs

Great Scott! The robots are coming for your job—and this time, they brought unit tests. Join Doc and Marty from the Software Future (Baruch and Leonid) as they race back in time to help you fight the machines using only your domain expertise, a well-structured prompt, and a pinch of Gherkin. This keynote is your survival guide for the AI age: how to close the intent-to-prompt chasm before it swallows your roadmap, how to weaponize the Intent Integrity Chain to steer AI output safely, and why the Art of the Possible is your most powerful resistance tool. Expect:

• Bad puns
• Good tests
• Wild demos

The machines may be fast. But with structure, constraint, and a little time travel, you’ll still be the one writing the future.

Decisions, decisions…

Want to find out more about and register for Arc of AI?

Once again, Arc of AI will take place from Monday, April 13 through Friday, April 16, with the workshop day taking place on Monday, and the main conference taking place on Tuesday, Wednesday, and Thursday.

Arc of AI tickets are BOGO!

From Arc of AI’s registration page:

You read that right! For each conference ticket you purchase, you get one free ticket. This applies only to conference tickets and not for workshops.

Categories
Artificial Intelligence Games Programming What I’m Up To

“Eternal Grind”: My agentically-coded homage to “Progress Quest”

Feeling nostalgic for the 2000s? Need a little amusement? I’ve got the agentically-coded thing you need: Eternal Grind!

Experience it now! Point your browser at accordionguy.github.io/eternal-grind/, then sit back and enjoy the adventure as the game plays itself for you. No effort required, and no time lost to the grind that other online role-playing games bring.

Screenshot of “Eternal Grind,” later on in the game.
A screenshot of Eternal Grind, later on in the game. Click to view at full size.

Once again, it’s here: accordionguy.github.io/eternal-grind/.

What’s Eternal Grind all about?

Eternal Grind is my version of Progress Quest, a parody of the popular 2000s game (and devourer of nerd lives) EverQuest. Unlike EverQuest, which was a multiplayer, Dungeons and Dragons-inspired role-playing game with a cluttered dashboard that required your full attention…

Screenshot of an Everquest game in progress.
A screenshot from EverQuest. Click to view at full size.

Progress Quest was a zero-player Dungeons and Dragons-inspired role-playing game that required no attention at all. It did keep one key aspect of EverQuest: the with a cluttered dashboard. Here’s a screenshot of the game in all its Windows XP glory:

Screenshot of Progress Quest, a Windows XP game made up entirely of list views and progress bars.
Progress Quest! Click to go to the official Progress Quest site.

Eternal Grind is my homage to Progress Quest. Like Progress Quest, it aims to be the ultimate “zero-player” RPG experience, providing all the dopamine of a legendary quest, but with absolutely none of the effort.

In the spirit of today’s best workflows, Eternal Grind automates the entire heroic journey, from slaying fantastical creatures like Literal Metaphors to hoarding  fabulous artifacts such as the Scissors of Regret.

The game automatically creates characters like Kevin from Accounting (a Low-Carb Orc and Spreadsheet Warrior by trade), after which your only job is to sit back and watch the progress bars fill. It’s a witty, Windows XP styled commentary on the nature of the “grind,” where the numbers always go up, the loot is perpetually absurd, and your lack of agency is the greatest feature of all.

Why are you still reading? Play it now! It’s here: accordionguy.github.io/eternal-grind/.

I built it with Zenflow

There are two notable differences between Eternal Grind and Progress Quest, the game to which it pays homage:

  1. While Progress Quest was a Windows-only desktop game, Eternal Grind is a single-page web game that runs on any device with a browser. Feel free to play it on your internet fridge!
  2. Zencoder logoProgress Quest was written the old-school way: using a programming language — namely, Delphi (Borland’s version of Pascal). Eternal Grind was written the new-school way: agentically, using Zencoder’s Zenflow AI coding tool.

That second point is an important one. Progress Quest was the product of traditional coding: the manual, instruction-based process where the developer acts as both architect and builder, meticulously and painstakingly writing instructions that specify how the program should do its work. Success depends on that developer’s ability to translate complex ideas into perfect syntax.

Eternal Grind is a different beast, since it’s the result of agentic coding, where the approach is intent instead of instruction. Instead of dictating the “how,” I provided a high-level specification — the “what” — to Zenflow, which can autonomously plan, write, and even self-correct the code.

(I’ll include the aforementioned specification at the end of this article.)

When using Zenflow to build Eternal Grind, I was no longer the contractor laying every brick. I was now the supervisor, providing the blueprints and overseeing an AI crew that did the bricklaying.

I plan to keep tweaking Eternal Grind using Zenflow. Be sure to visit its page often!

One more time: Eternal Grind is at accordionguy.github.io/eternal-grind/.

The specification

Eternal Grind started with a specification that I wrote into file named spec.md.  This file served as the definitive “source of truth” that described the kind of application I wanted created. While traditional specs are often treated as a “nice-to-have” for human developers, AI agents needs such a spec to act as a “North Star” as well as to keep them from developing the wrong thing.

By clearly defining the application’s logic, layout, and data in a structured format, I provided Zenflow with the basic context for building Eternal Grind. It turns a vague, hand-wavey request into a structured mission, ensuring that the code generated not only just works, but also provides the application I expected, working in the way I expected.

Here’s the complete specification file I initially wrote:

# Functional Specification: Eternal Grind (ZPRPG)

## 1. Project Overview
"Eternal Grind" is a "Zero-Player RPG" (ZPRPG) inspired by the classic parody *Progress Quest*. The game automates all traditional RPG elements—questing, combat, looting, and leveling. The user's role is purely observational.

---

## 2. UI Layout (Three-Column Dashboard)
The application shall use a fixed-height, full-width dashboard layout using Flexbox or Grid.

### A. Character Sheet (Left Column - 25% Width)
* **Identity:** Displays Character Name (from `NAMES`), Level, Race, and Class.
* **Stats Table:** A vertical list of numerical values for the 10 core stats (e.g., Strength, Existential Dread).
* **Equipment:** A list of 6-10 equipment slots showing absurd gear.
* **Spells/Abilities:** A scrolling list of learned "skills" that grows upon leveling up.

### B. The Engine of Progress (Center Column - 50% Width)
* **Location Header:** Displays the current location from the `LOCATIONS` list.
* **Primary Task Bar:** A large progress bar indicating the current action (e.g., "Contemplating the void").
* **Plot Bar:** A slower-moving bar tracking progress toward the next "Act."
* **Experience Bar:** A bar tracking progress toward the next Level.
* **Portrait:** A central area for a static character icon or simple CSS animation.

### C. Data Feed (Right Column - 25% Width)
* **Inventory (Top Half):** A scrolling list of items collected. Maximum capacity: 15 items.
* **Quest Log (Bottom Half):** A vertical scrolling log of events. It must automatically scroll to the bottom as new lines are appended.

---

## 3. Core Mechanics & Logic

### 3.1 Initialization
When the application starts:
1.  **Name Selection:** A name is chosen randomly from the `NAMES` list and remains permanent.
2.  **Character Build:** A `RACE` and `CLASS` are randomly assigned.
3.  **Starting Stats:** Each stat in the `STATS` list is assigned a random base value between 3 and 18.

### 3.2 The Game Loop
The application runs on a continuous timed loop:
1.  **Questing:** The "Task Bar" fills over a period of 3–8 seconds.
2.  **Completion:** Once the bar hits 100%:
    * A random **Monster** is "defeated."
    * A random **Item** (Adjective + Noun) is added to the Inventory.
    * A line is added to the **Quest Log** (e.g., "Executed a Low-Level Bugbear. Found: Rusty Sock of Mystery").
    * The **Experience Bar** increments.
3.  **Market Mode:** When the Inventory reaches 15 items:
    * The current task changes to "Heading to market to sell junk."
    * After a short delay, the Inventory is cleared and the character returns to questing.
4.  **Leveling Up:** When the Experience Bar reaches 100%:
    * The Character Level increments.
    * A random **Stat** increases by 1.
    * A new **Spell** is randomly selected and added to the spell list.
    * The Experience Bar resets.

---

## 4. Technical Requirements
* **State:** The application must maintain a state object containing the character's profile, stats, inventory list, and log history.
* **Styling:** A "Retro Win95" or "Classic MMO" aesthetic with high-contrast borders.
* **Performance:** The log should prune entries older than 100 lines to maintain performance.

---

## 5. Data Appendix

### Character Names
* Kevin from Accounting, Sir Tap-A-Lot, The Great Barnaby, User_772, Mistake #4, Sir Not-Appearing-In-This-Game, A Literal Bag of Flour, Lord Helvetica, Chadwick the Unready, Karen of the Suburbs, Glitchy McGlitchface, The Placeholder, Grommet the Slightly Agitated, Barb the Librarian, Sir Sells-Everything, Kyle the Monster Energy Enthusiast, Grandmaster Procrastinator, The Unpaid Intern, Sir Buffering..., Standard Hero 01.

### Races
* Sentient Toaster, Depressed Elf, Low-Carb Orc, Middle-Management Dwarf, Glitch in the Matrix, Half-Empty Human, Sentimental Slime, Vague Shadow, Procrastinating Pixie, Bureaucratic Beholder, Existential Ghost.

### Classes
* Spreadsheet Warrior, Chronic Procrastinator, Underpaid Mage, Professional Mourner, Existentialist Rogue, Lunch Knight, Intermittent Faster, Coffee Warlock, Passive-Aggressive Paladin, Technical Support Druid, Tax Accountant.

### Tasks
* Debating a fence post, Polishing a rusty nail, Contemplating the void, Waiting for a sign, Filing a 1040-EZ, Staring into the middle distance, Organizing a sock drawer, Explaining the internet to a rock, Searching for a lost remote, Counting ceiling tiles, Simulating a personality, Buffing out a scratch in reality.

### Locations
* The Forest of Mild Inconvenience, The Cave of Echoing Sighs, Downtown Boredom, The Desert of Dry Humor, Mount Mediocrity, The Swamps of 'I'll Do It Tomorrow', The Suburbs of Despair.

### Item Adjectives
* Dull, Polished, Forbidden, Rusty, Lamentable, Insignificant, Glowing, Slightly Damp, Overpriced, Mediocre, Legendary-ish.

### Item Nouns
* Scissors of Regret, Pebble of Mediocrity, Scone of Power, Lint of Destiny, Paperclip of Hope, Broken Twig, Expired Coupon, Sock of Mystery, Unfinished Novel, Jar of Pickled Thoughts.

### Monsters
* A Literal Metaphor, The Concept of Ennui, A Low-Level Bugbear, An Imaginary Friend, A Confused Salesman, A Dust Bunny of Doom, The Ghost of a Dead Pixel, A Sentient Terms of Service Agreement.

### Spells
* Aggressive Sighing, Metaphysical Poke, Summon Minor Annoyance, Greater Procrastination, Flash of Inadequacy, Power Word: 'Whatever', Cloud of Confusion, Internal Monologue.

### Stats
* Strength, Constitution, Dexterity, Intelligence, Wisdom, Charisma, Patience, Luck, Caffeine Level, Existential Dread.

Zenflow generated the application, and I also had it use a different agent to review its own code.

I ran the application, saw things I wanted changed, and then specified those changes:

Screenshot of one of my interactions with Zenflow while building Eternal Grind
One of my change requests in Zenflow. Click to view at full size.

Zenflow made the changes, then I had the review agent review those changes. This process of refinement continued for a couple more steps, and the result is the game located at accordionguy.github.io/eternal-grind/.

As I mentioned before, Eternal Grind is a work in progress. I’ll continue adding tweaks and improvements using Zenflow. Watch this space!

Find out more

 

Categories
Artificial Intelligence Conferences Programming What I’m Up To

The Arc of AI conference’s workshop day: Monday, April 13, 2026

From April 13th through 16th — and a couple of days before, because it’s in Austin — I’m going to be at the Arc of AI conference! Over the next little while, I’m going to be posting articles about Arc of AI, in case you’re wondering what the conference is about and whether you should go.

In this article, I’ll talk about the workshop day and one of the workshops in particular.

Monday, April 13: The workshop day

Screenshot of the workshops schedule for Arc of AI’s workshop day.
Click to see the workshops at full size.

Prior to the main conference days (Tuesday, April 14 through Thursday, April 16), Arc of AI will hold its Workshop Day on Monday, April 13, where they’ll have six AI workshops:

  • Fundamentals of Software Engineering In the age of AI (Dan Vega and Nathaniel Schutta)
  • Building a Production-Grade RAG Pipeline (Wesley Reisz)
  • AI-Driven API Design (Mike Amundsen)
  • Creating AI Assisted Applications Using LangChain4j (Venkat Subramaniam)
  • Developing AI Applications with Agents, Rag, and MCP using Python (Brent Laster)
  • Tech Leadership in the Time of AI (Brian Sletten)

The Fundamentals of Software Engineering in the Age of AI workshop

One of the workshops I’m interested in is Nathaniel Schutta’s and Dan Vega’s Fundamentals of Software Engineering in the age of AI, which will be based on their recently-published (November 2025) O’Reilly book, Fundamentals of Software Engineering, but with the application of AI.

Here’s an excerpt from their workshop’s abstract:

This intensive workshop bridges the critical gap between what early-career developers learn in formal education and what they need to thrive in professional environments where human expertise and artificial intelligence increasingly collaborate. Based on our book “Fundamentals of Software Engineering,” we guide participants through a comprehensive journey from programmer to well-rounded software engineer equipped to leverage AI tools effectively while maintaining engineering fundamentals.

Participants will develop both technical capabilities and professional skills that remain relevant regardless of changing languages, frameworks, and AI capabilities. Through a balanced mix of conceptual teaching, collaborative discussions, and hands-on exercises with both traditional and AI-assisted approaches, attendees will work on realistic scenarios that reinforce practical application of these fundamental principles while developing discernment about when and how to integrate AI tools into their workflow.

Learnings:

  • Understanding the programmer to engineer transition and mindset shift
  • Developing advanced code reading techniques and comprehension strategies
  • Crafting maintainable, readable code that communicates intent
  • Applying software modeling concepts to visualize and plan complex systems
  • Implementing comprehensive automated testing strategies
  • Effective techniques for working with legacy codebases and existing systems

Benefits:

Students will understand the concepts and how to apply them right now cutting through the hype surrounding AI. With practical tips and guidance, they can jumpstart their use of AI across the software development lifecycle.

Who should attend:

Primarily developers and architects but ultimately anyone that’s struggling to understand how to apply AI to their world today while avoiding the pitfalls and rabbit holes.

I’m intrigued by this workshop, as it’s about the application of AI tools to the way software is built, which is pretty new turf for all of us. When I learned software development, there were already plenty of lessons from decades of developers’ experiences, and in my career, I and the rest of the industry picked up a couple decades’ more tips and tricks. But all that learning is from the “before times.” Right now, we’re not even five years into the post-ChatGPT era, and we’re only beginning to figure out how to write applications in the era of vibe coding (and remember, Andrej Karpathy coined the term barley over a year ago).

Since the workshop is based on the book, this video might give you an idea of what it’ll be like:

Want to find out more about and register for Arc of AI?

Once again, Arc of AI will take place from Monday, April 13 through Friday, April 16, with the workshop day taking place on Monday, and the main conference taking place on Tuesday, Wednesday, and Thursday.

Arc of AI tickets are BOGO!

From Arc of AI’s registration page:

You read that right! For each conference ticket you purchase, you get one free ticket. This applies only to conference tickets and not for workshops.

 

Categories
Picdump

Saturday picdump for Saturday, March 14

Happy Saturday, everyone! Here on Global Nerdy, Saturday means that it’s time for another “picdump” — the weekly assortment of amusing or interesting pictures, comics, and memes I found over the past week. Share and enjoy!



Screenshot
Screenshot



Screenshot
Screenshot


why-not-store-everything-in-cache

hair-dryer

Screenshot
Screenshot


649415445_1350333820448293_4115571728292986423_n

the-hour-we-lost

if-you-can-lean

comments-section

why-do-you-show-up-to-work-irritated

650811480_1351792610302414_4609301814111761059_n

how-to-not-use-ai

grok

we-made-social-media

650216265_1351286513686357_1371194668056750165_n

spot-the-wap

IMG_1532

IMG_1271

650082694_1349739697174372_1861492054552790623_n

Screenshot
Screenshot


you-are-here

650050538_1350175570464118_6527345586530996700_n

og_infinity_stones

649336002_1349928853822123_940106382801534275_n

650138061_1351647383650270_7355935145823925561_n

650500987_1350974273717581_2721688371741210720_n

649323603_1350333887114953_1839934216097563918_n

650340698_1350659930415682_487033684383132965_n

bad-screen-good-screen

Screenshot
Screenshot


nuclear-war

Screenshot
Screenshot


real_life_has_become_rick_and_morty

652667262_1351852960296379_7054292108599786352_n

650600843_1351223067026035_3268709820066176183_n

648592298_1351223017026040_2987059082564608342_n

its-been-forever

Screenshot
Screenshot


nanny-wanted

extra-shift

650123731_1350032403811768_48605464345538844_n

comment-section

651212966_1351571696991172_1303932270148797433_n

local-llms

chipotleGPT

were-cooked-bro

Screenshot
Screenshot


651363136_1350140457134296_6553703424557869044_n

where-do-you-see-yourself-in-five-years

650230914_1350974203717588_2427671003887412673_n

sloprank.io_

IMG_1508

ats-dashboard

online-death-certificate-request

650747825_1350304050451270_6410710245789885711_n

changing_jobs

Screenshot
Screenshot


IMG_1512

so-much-to-do

free-ddr5-ram

keygen

650104046_1349834553831553_3778352494133341844_n

Screenshot
Screenshot


650411371_1350850490396626_593867738032850239_n

650136953_1351078053707203_165746247321824773_n

647212548_26040696855581359_6949754323448152088_n

651184567_1350933073721701_837546210975592430_n

fake-idioms

handle-it

651079924_1351608776987464_852844420925762066_n

targeted-ads

claude-max-and-code-review

alex_karp

IMG_9643

positive-attitude

Screenshot
Screenshot

Categories
Artificial Intelligence Conferences

Notes from Advantage, part 4 — Dave Parry: Shift to Agentic Software Engineering

Last week, Anitra and I attended both the Dev/Nexus conference and its companion conference, Advantage, an AI conference for CTOs, CIOs, VPs of Engineering, and other technical lead-types, which took place the day before Dev/Nexus. My thanks to Pratik Patel for the guest passes to both conferences!

I took copious notes and photos of all the Advantage talks and will be posting summaries here. This set of notes is from the fourth talk, Shift to Agentic Software Engineering, presented by Dave Parry.

Here’s Dave’s bio:

David Parry is an accomplished Director of Architecture with over 20 years of experience in Software Development. It all began in 1996 when he discovered the fascinating world of programming, with a particular focus on Java applets. Throughout his illustrious career, David Parry has been involved in various noteworthy projects. He has successfully built and implemented content management systems for a wide range of clients, including the esteemed Johny Walker and its renowned keepwalking.com. Additionally, as a consultant at a Big 4 firm, David played a pivotal role in solving critical issues for numerous customers, demonstrating his expertise in handling complex and high-traffic web platforms. Never one to shy away from innovation, David Parry has expanded his skills to work on cutting-edge technologies such as mobile and embedded Android TV systems. Leveraging his expertise, he has delivered top-notch streaming services to customers, ensuring they have an exceptional viewing experience. Currently, David holds the position of Developer Advocate and Consultant overseeing strategic planning and execution of architectural designs for customers. With a deep understanding of software development principles and extensive experience in Java programming, he excels at providing valuable insights and guidance to his team. Having witnessed the evolution of Java development from its early days to its current state, David Parry’s wealth of experience and strategic perspective, combined with his consulting work at a Big 4 firm, make him an invaluable asset in any project or organization he is a part of.

And here’s the abstract of his talk:

AI is redefining how engineering organizations operate, shifting from traditional development to agentic development, where intelligent, context-aware agents partner with teams to drive measurable business outcomes. This presentation gives leaders a clear framework for understanding how agentic development improves cycle time, reduces operational risk, enhances quality, and scales organizational capacity without adding headcount. We will examine how to move beyond pilots, achieve meaningful adoption, embed governance and security controls, and connect engineering effort directly to enterprise KPIs. Leaders will leave with a strategic roadmap for guiding their organizations through this transformation with clarity, confidence, and control.

My notes from Dave’s talk are below.


The shift from AI-assisted to agentic is real, and most organizations aren’t ready

Dave opened by drawing a line between two distinct eras of AI in software development. The first era, the era of AI-assisted coding/the GitHub Copilot model, still has a human in the loop at every step. A developer reviews suggestions, accepts or rejects them, and retains full decision-making authority. This is the model most development teams have actually adopted, and it‘s valuable. The second era, agentic software engineering, is something categorically different: autonomous systems that execute multi-step workflows without continuous human supervision.

Dave was candid that most organizations are still figuring out how to use AI-assisted tools well, even as the industry conversation has moved on to agents. The gap between where the hype is and where most teams actually are is significant, and leaders who try to leapfrog directly to full autonomy without establishing the right foundations tend to end up with agents that are expensive, unpredictable, and politically toxic inside the engineering organization. The smarter path, in Dave‘s experience, is to build the scaffolding — governance, measurement, structured experimentation — before letting agents loose on anything consequential.

Governance can‘t be bolted on after the fact

The governance message in Dave‘s talk was clear: security and access controls must be architected into agentic systems from the beginning, not added as an afterthought once the agent is already running. He illustrated this with a client story about a company whose repositories were so strictly siloed that individual developers weren‘t even allowed to know other repos existed, let alone access them. An agent given broad permissions in that environment would immediately violate carefully constructed security boundaries that humans had been respecting for years, simply because nobody thought to encode those constraints into the agent‘s operating parameters.

The practical implication is that every constraint your human engineers operate under (such as access controls, data isolation, permission scoping) needs to be explicitly defined for any agent working in the same environment. Agents don‘t have professional judgment or social awareness; they will access whatever they‘re technically permitted to access. If you onboard a new human developer, you scope their access carefully before they write a single line of code. Agents require the same rigor. Dave‘s recommendation was to look for frameworks that make these governance constraints first-class concepts rather than optional configurations, and to be deeply skeptical of any agentic solution that treats security as something you layer on later.

Enterprise-readiness also extends to the technology choices themselves. Dave pushed back against agentic frameworks built in languages or runtimes that don‘t fit naturally into enterprise operational environments. A security team asked to approve an agent that spins up an npx process that re-downloads dependencies on every run is going to say no…and they should! The same agent behavior built on Spring Boot, running in a container with Prometheus observability already wired in, is a fundamentally different conversation.

Measure everything! Agents aren’t self-evidently valuable

One of Dave‘s most pragmatic points was that the business case for any given agent needs to be proven, not assumed. The pressure from above to “do AI” is real, but implementing an agent that costs more in compute and maintenance than it would cost a developer to do the same task manually is not a win — it‘s a liability that will eventually get noticed and used to discredit the entire program. Leaders who can‘t quantify what their agents are actually delivering are in a precarious position when budget scrutiny arrives.

His recommendation was to tie every agent deployment to concrete, measurable KPIs from the start. For a PR risk agent, the relevant metrics might include change failure rate, time to production, and whether bug rates are actually going down or inadvertently going up as junior developers blindly accept AI suggestions. The five-star anecdote was a useful cautionary note: some teams have discovered that their agents were actively introducing more defects than they prevented, precisely because they hadn‘t built in the measurement infrastructure to detect it early.

Dave also pushed back against the proof-of-concept mentality that treats agent work as inherently experimental. The POC era, in his view, is over. Organizations that frame every agent initiative as “let‘s see if this works” create the conditions for naysayers to kill it at the first sign of friction. His preferred framing is to pick a small, low-risk pilot, commit to shipping it to production, measure it rigorously, and use that concrete success to build momentum for the next one. Owning the conversation with data is the only reliable way to keep agentic programs alive long enough to deliver real compounding value.

Bring your existing developers into the agentic transition; don‘t route around them

A consistent thread throughout Dave‘s talk was that agentic AI is not a replacement for experienced engineers, but an amplifier of their knowledge. That amplification only works if those engineers are inside the tent. Developers who feel threatened by agents will find reasons for them to fail, and frankly, they‘ll often be right, because agents built without deep domain knowledge embedded in their prompts and tools tend to produce plausible-looking but subtly wrong outputs. The engineers who know where the bodies are buried in your codebase are exactly the people who should be shaping how your agents operate.

Dave‘s specific recommendation was that when outside expertise comes in to help stand up an agentic program, that expertise should be focused on upskilling the existing team rather than doing the work for them. An external consultant who delivers a finished agent and walks away leaves the organization with something it doesn‘t fully understand and can‘t maintain or evolve. An expert who works alongside the existing team, transfers knowledge, and helps them build the verification and governance capabilities they need to operate agents independently is creating something durable.

Dave made the point that custom MCP servers are one of the highest-leverage things an organization‘s own developers can build, because that‘s where domain-specific knowledge gets embedded in a form that agents can reliably use. A generic MCP that connects to a database and lets the LLM figure out the schema from scratch on every query is both expensive in tokens and fragile in output. A purpose-built MCP that encodes exactly what that database contains, how to query it correctly, and what the results mean — written by developers who actually know the system — is the kind of deterministic grounding that makes agentic systems genuinely trustworthy in production.