|
Sr. Content Developer at Microsoft, working remotely in PA, TechBash conference organizer, former Microsoft MVP, Husband, Dad and Geek.
|
What AI thinks a DBA does all day is a calm, orderly, deeply reassuring Tuesday. It has backups, documentation, proactive health checks, and apparently a door that closes. I asked. Then I wrote down my actual Tuesday. Its day has six items. Mine has thirteen.
I asked it to describe a typical day for a database administrator. It answered instantly, with the confidence of something that has never been cc’d.
| What AI Thinks Happens | What Actually Happens |
|---|---|
| 9:00. Review overnight backup jobs and monitoring alerts. | 9:04. Somebody asks whether the database is slow. The database is not slow. Their Wi-Fi is slow. They remain unconvinced. Nobody has ever escalated a router. |
| 10:00. Analyse slow-running queries and optimise execution plans. | 9:20. A developer needs sysadmin. Just temporarily. Just for one thing. The one thing is all of them.
9:45. Meeting about a project that will not happen. We build a timeline anyway. By 10:25 it has a name, three owners and no future. |
| 11:30. Collaborate with development teams on schema design. | 10:30. The printer.
11:15. Somebody committed an |
| 13:00. Capacity planning and index maintenance. | 12:00. A vendor asks me to disable a security setting. Their thirty-eight-page PDF calls this a documented requirement. Documented means typed.
13:30. Someone asks if I can just pull a quick report. Just means they have estimated my effort at zero. Quick means they want it before they finish asking. I ask what it is for. They say they will know when they see it. |
| 15:00. Update documentation and share knowledge with the team. | 14:00. Half an hour of genuine, uninterrupted database work. I look up and it is 14:12.
14:15. I raise a sandwich. A message arrives: “quick question.” It is a project wearing a question mark. I lower the sandwich. 15:00. I open the documentation to update it. The last edit is mine. It is from 2019. |
| 16:30. Proactive health checks and performance baselining. | 15:40. The same alert that has fired every Tuesday since 2021 fires again. Nobody will turn it off. Turning it off would mean knowing what it does. The alert has tenure.
16:50. I finally sit down to tune the query I have wanted to tune since Thursday. 16:53. The printer. It has sensed optimism. |
The left column has a rhythm. Tasks begin. Tasks end. Lunch appears to exist. The right column has an arrival rate. The sandwich is still on the desk.
One item did line up. Its 10:00 and my 16:50 are the same task. It allowed ninety minutes. I got three.
I would like to work in the left column. I meet every documented requirement. Please send me the application link.
I don’t know why. I never have. I’ve worked with databases for twenty years, and in every organisation there comes a quiet ceremonial moment when the printer becomes mine. No ticket assigns it. No manager announces it. The knowledge passes down like folklore, except folklore rarely needs a magenta cartridge. At one place I was shown the printer during the tour on my first morning. Nobody said what it was for. They just slowed down as we passed it.
I don’t print. There is a magenta cartridge in my laptop bag. I bought it myself. It was faster than explaining. It does not fit the current printer. I have not taken it out.
It wasn’t wrong, exactly. It was accurate to the job description. Every line of it appears in the posting that was written when the role was created, and in the training material, and in the certification syllabus, and in roughly four hundred thousand blog posts, several of them mine. I helped write the version of this job that does not have a printer in it. It read me. It believed me.
Almost none of my Tuesday is written down anywhere. Nobody wrote down that the printer is mine. Nobody wrote down which server you can restart quietly and which one has a person attached to it. Nobody wrote down that when a certain team says urgent they mean Thursday, and when a different team says whenever you get a chance they mean now. Knowing the difference is worth more than any certification I hold, and I hold a decorative number of them. That isn’t database administration. It is translation between local dialects of panic.
None of it’s in the training data, because none of it’s written down. And none of it’s written down because everybody already knows. Everybody means the five people who have been there long enough to stop asking why. The five people are the documentation. They have not been backed up.
Nobody writes down the parts of the job they’re still hoping are temporary. I have been hoping since 2006.
So I pasted my actual Tuesday in and asked it to write the job posting.
It did, immediately. It did not ask what it was for. The Wi-Fi argument became stakeholder communication. The sysadmin request became access governance. The meeting about the project that will not happen became cross-functional planning. The documentation I have not touched since 2019 became ongoing knowledge management. Every line was accurate. Not one of them was recognisable.
The printer did not appear. I looked twice. My Tuesday went in with a printer in it and came out clean.
It listed eight requirements. I meet six of them. They came from my own Tuesday.
Somebody’s real Tuesday went through this years before mine. What came out is the left column. The job description is not a fantasy. It is a real day with the printer taken out.
You could read that as good news. The machine can’t see the real job, so the real job is safe.
If your work is invisible to something that has read almost everything, it’s invisible to your HR system too. They’re all reading the same documented version of you. No appraisal has ever mentioned the printer. The documented version of you is the one applying for your next job. It interviews better.
The machine isn’t misunderstanding your job. It’s repeating back the only version of it anybody wrote down.
So write down the expensive parts. The judgement calls, the team translations, the seven undocumented things that only work because you know they do. Leave the printer for the oral tradition.
I went back and asked whether a DBA is ever responsible for office printers. I described it. I typed the question standing at the printer.
It said no. It said that would fall outside the typical scope of the role.
It sounded so certain. I didn’t correct it. One of us should have boundaries, and apparently it is the machine.
This is the same argument sitting underneath all thirty essays in my book AI: Nobody’s in There. But we’re still in here. All thirty are free to read at pinaldave.com, and the book is available in paperback, Kindle and audiobook on Amazon.
The machine did not get my job wrong. It read the job description. The funny part is everything the job description forgot. The dangerous part is that everybody else reads the same one. The printer is not mine.
Reference: Pinal Dave (https://blog.sqlauthority.com/), AI and DBA, X
First appeared on What AI Thinks a DBA Does All Day
A field guide to .NET data formats
9 minutes by Bipin Joshi
Different data formats each have their own query language suited to their structure. CSV relies on parsing libraries like CsvHelper before handing off to LINQ, while SQL remains the most mature standalone query language for relational data. XML uses XPath or LINQ to XML for tree traversal, JSON favors typed deserialization, and binary formats skip querying entirely. Most formats are just a path to in-memory objects, where LINQ takes over.
Rider's .NET Debugger in VS Code and Cursor? Yes, Really!
sponsored by Jetbrains
The ReSharper extension brings the Rider debugger engine to VS Code, Cursor, and other compatible editors. Enjoy full breakpoint support, watch expressions, step navigation, launch and attach for .NET, .NET Framework, Mono, and web apps, plus unit‑test debugging – and your existing launch.json configs are picked up automatically.
Building a distributed job scheduler
11 minutes by Aaron Stannard
Akka.Cluster.Sharding and cluster routers both fall short for long-running jobs: sharding rebalances and kills in-progress work, while routers lack memory and recovery. A better approach uses a persistent singleton actor as a central scheduler that tracks job sizes, picks the least-loaded node, and never moves work already in flight. When nodes join or crash, only queued jobs are redistributed, leaving running jobs untouched.
Faster blob copying in azure blob storage
6 minutes by Mark Heath
Copying blobs between Azure storage accounts can be done in three ways, with speed varying wildly. Downloading and reuploading is the slowest, taking over an hour for a 10GB file, as it is limited by your local connection. A server-side copy is faster but still slow across accounts. The fastest approach splits the file into blocks and copies them in parallel entirely within Azure, completing the same 10GB copy in under 10 seconds.
Introducing Postie
6 minutes by Andrew McLachlan
Postie is a free, open source library for ASP.NET Core that removes boilerplate from the CQRS pattern. Instead of writing repetitive handler lambdas for every endpoint, you map commands and queries directly to routes in a single line. It works with its own built-in mediator or with MediatR, and handles common HTTP conventions like 201 responses and 404s automatically.
Parsing IP addresses in C# at crazy speeds
3 minutes by Daniel Lemire
This article shows how to parse IP addresses in C# using AVX-512, a newer CPU instruction set. It uses masked loads to handle short strings safely, then locates dots and validates the address with vector math. The result is about three times faster than the standard library method, reaching roughly 71 million addresses per second on modern hardware.
And the most popular article from the last issue was:
Passkeys in ASP.NET Core 10
11 minutes by Adrian Bailador
ASP.NET Core Identity in .NET 10 ships built-in WebAuthn/FIDO2 passkey support. Here's how registration and sign-in actually work under the hood, what MakePasskeyCreationOptionsAsync and PasskeySignInAsync do for you, and where the implementation deliberately stops short of a full WebAuthn library.
Cut AI Coding Token Costs by up to 36% with Sonar Vortex
sponsored by Sonar
Sonar Vortex operates inside your AI coding agent’s inner reasoning loop, supplying deep architectural context before code is written, then verifying the output in real time. In testing, software defects dropped by 92% and token consumption decreased. Build safer code while spending less on LLM calls.
REST API security best practices in ASP.NET Core
19 minutes by Anton Martyniuk
Building secure ASP.NET Core APIs comes down to consistent habits across the whole codebase. Anton covers 18 practices, from enforcing HTTPS and validating JWT tokens properly to preventing SQL injection with EF Core and capping request sizes. Key themes include treating all input as hostile, using policies for authorization, storing secrets outside source control, and keeping dependencies patched. No single fix is enough, but starting with HTTPS, token validation, and input handling closes the biggest gaps fast.
Cache invalidation in .NET
5 minutes by Yohan Malshika
Caching speeds up apps but causes problems when data changes. Cache invalidation solves this by removing or updating stale cache entries after database changes. The Cache-Aside pattern handles this well: read from cache first, and after any update or delete, remove the related cache entry so the next request fetches fresh data. Adding a TTL gives extra protection, but explicit invalidation is still needed when accuracy matters.
The role of ActivitySource in OpenTelemetry for .NET
5 minutes by Bart Wullems
ActivitySource is the entry point into .NET's native tracing story, and OpenTelemetry is just one listener on top of it. The name is the contract; get that right, register it with AddSource, and you can create as many instances of it as your codebase's structure naturally calls for. Don't let the fear of "duplicate instances" push you into DI ceremony you don't need.
EF Core interceptors: The complete guide
16 minutes by Mukesh Murugan
EF Core interceptors act as middleware for database calls, letting you observe, modify, or cancel operations like saving changes or running SQL without touching the calling code. The two most useful types are ISaveChangesInterceptor for audit fields and soft deletes, and IDbCommandInterceptor for SQL-level work like logging slow queries. To inject a scoped service like the current user, register the interceptor as scoped and resolve it through the AddDbContext service provider overload.
And the most popular article from the last issue was:
Anthropic’s best AI model struggles to attract users as cheaper tools thrive
A few interesting numbers in this FT story gathered from "people with knowledge of the matter":This article also introduced me to the Ramp AI index, which uses billing data from 70,000 Ramp credit card using companies to estimate model adoption.
Here's Ramp's breakdown of Anthropic model spend for July 2026, which looks reasonable given that Opus 5 was only released on July 24th, and supports the idea that Fable's cost has made it a less popular model:
Via Hacker News
Tags: ai, openai, generative-ai, llms, anthropic, claude, claude-mythos-fable
Get caught up on the latest technology and startup news from the past week. Here are the most popular stories on GeekWire for the week of Aug. 16, 2026.
Sign up to receive these updates every Sunday in your inbox by subscribing to our GeekWire Weekly email newsletter.
404 Media put an Apple AirTag in a rare book and tracked it to an Amazon warehouse in Las Vegas, where a team cuts the bindings off books and scans the pages. The reporting method has roots in a Seattle non-profit group. Read More … Read More
A year after GeekWire asked where Seattle’s superstar AI startups were, two local venture capitalists offer an answer: they’re here, but they build rockets, geothermal plants and autonomous off-road vehicles rather than software. A guest opinion from Ascend’s Nate Bek and Plug and Play’s Ben Eisinger. Read More … Read More
Qualtrics laid off employees across Seattle, Provo and international offices Wednesday, three months after closing its $6.75 billion acquisition of Press Ganey Forsta. The company declined to disclose how many jobs were cut, but a WARN notice reviewed by GeekWire indicates at least 50 cuts were made at its Seattle headquarters. Read More … Read More
After the U.S. Navy abruptly canceled a key unmanned warship program, defense giant Anduril quietly vacated its shipyard site on Seattle’s Lake Washington Ship Canal—even as its broader regional footprint continues to surge toward 1,000 local engineers. Read More … Read More
Expedia Group is parting ways with at least eight vice presidents and senior vice presidents as it reorganizes its product and technology organization around AI, according to an internal memo obtained by GeekWire. The memo from the company’s product and technology chiefs also names five leaders who were promoted. Read More … Read More
GitHub was down for more than three hours Monday morning, breaking the website, code review tools, automated build systems and Copilot. It’s the latest reliability problem for a platform whose capacity planning has been outrun by AI coding tools. Read More … Read More
An AWS director who ran the launch of Oracle Database@AWS has joined Oracle. Also in this edition: Seattle Children’s names a Stanford informatics leader as CIO, Zillow fills a newly created chief legal and policy role and a former Convoy executive is Uber Freight’s new product chief. Read More … Read More
GeekWire toured Anduril’s unmarked Bellevue office, where the defense company is building augmented reality glasses, digital night vision and edge computing hardware for the U.S. military. Senior vice president Tom Keane says the regional engineering workforce could easily grow from 560 people to 1,000 in the not-too-distant future. Read More … Read More
Endurance Energy is running full-tilt in its cavernous new Seattle facility as it prepares to deploy a 100-kilowatt generator to an oceanic volcano off the coast of Oregon. The company is betting that the technology can one day electrify entire cities. We got a look inside. Read More … Read More
The King County Library System opened a $5.2 million sorting installation in Renton that handles the 25,000 items moving through its 50 libraries each day. Robotic arms destack totes every 15 seconds and route materials into 186 delivery chutes. Read More … Read More
Jen Abel is the co-founder of JJellyfish and GM of enterprise sales at State Affairs. She is widely regarded as one of the sharpest practitioners in enterprise sales, and for that reason, this is her third visit to the podcast. In our first conversation we went deep on founder-led sales; in our second we mapped the $1M–$10M playbook. This time we do something I’ve never seen on another podcast: walk step by step through the full enterprise sales cycle. Most people think it’s five steps. Jen shows it’s closer to 15.
In our in-depth conversation, we discuss:
1. Why the standard five-stage CRM pipeline is a forecasting tool, not a sales process, and what the real 15-step cycle looks like
2. The “pincer model” for landing the first meeting at the executive and N-minus-one level simultaneously
3. How to craft a two-to-three-sentence message around giving them “alpha”
4. How to run an intro call that extracts maximum intelligence before you ever show a demo
5. The two-to-three-day pilot structure, how to define success jointly, and when to charge for a longer pilot versus giving it away
6. Navigating pricing, procurement, redline negotiations, and the final signature without losing momentum
—
Brought to you by:
WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more
Mercury—Radically different banking, now with Command
—
Where to find Jen Abel:
• LinkedIn: https://www.linkedin.com/in/earlystagesales
• Website: https://www.jjellyfish.com
—
Where to find Lenny:
• Newsletter: https://www.lennysnewsletter.com
• X: https://twitter.com/lennysan
• LinkedIn: https://www.linkedin.com/in/lennyrachitsky/
—
In this episode, we cover:
(00:00) Introduction
(02:49) Why Jen is giving away her enterprise sales playbook
(05:54) Example: selling AI legal tech to SpaceX
(09:11) Step 1: Landing the meeting
(18:44) Step 2: Running the intro call
(30:00) Step 3: Running a follow-up call
(36:30) Step 4: Prepping the pitch/frame for demo
(38:36) Step 5: Running the demo
(46:45) Step 6: Post-demo debrief
(48:59) Step 7 through 9: Preparing for and running the pilot
(01:03:44) Why the standard five-stage CRM pipeline doesn’t work
(01:04:49) Step 10: Post-pilot session
(01:07:48) What a healthy enterprise win rate looks like
(01:11:45) Steps 11 through 14: Navigating procurement and getting the signature
(01:18:01) How to think about expansion
(01:19:18) How buyers can say no without wasting everyone’s time
(01:21:53) Why enterprise sales is a lot like product management
(01:22:27) Closing thoughts
—
Referenced:
• State Affairs: https://stateaffairs.com
• The ultimate guide to founder-led sales | Jen Abel (co-founder of JJELLYFISH): https://www.lennysnewsletter.com/p/master-founder-led-sales-jen-abel
• “Sell the alpha, not the feature”: The enterprise sales playbook for $1M to $10M ARR | Jen Abel: https://www.lennysnewsletter.com/p/the-enterprise-sales-playbook-1m-to-10m-arr
• SpaceX: https://www.spacex.com
• Lemlist: https://www.lemlist.com
• Palantir: https://www.palantir.com
• Jason Lemkin on X: https://x.com/jasonlk
• We replaced our sales team with 20 AI agents—here’s what happened | Jason Lemkin (SaaStr): https://www.lennysnewsletter.com/p/we-replaced-our-sales-team-with-20-ai-agents
• Building a world-class sales org | Jason Lemkin (SaaStr): https://www.lennysnewsletter.com/p/building-a-world-class-sales-org
• Careers at State Affairs: https://stateaffairs.com/careers
—
Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email podcast@lennyrachitsky.com.
—
Lenny may be an investor in the companies discussed.
Log in
