Burn It Down, Rise of the Digital Twin
Dispatches from Mirrorworld on building for the individual
If I had a digital twin, what would she do?
Today, Andii2 would scour the web for interesting companies with missions I’d love and write compelling outreaches to CEOs and investors.
She’d read, skim, skip through the ocean of information and news each day floating up the things I might find interesting, leaving me the fun bit of diving down rabbit holes from the messy collection of newsletters, blogs, and X ramblings I typically pour through every day for a few hours.
Taking away the tedium, summarizing the need to know, and leaving the treasures for me to discover and dig through.
I could ask her to handle those things I keep putting off - upping my renter’s insurance policy, or some mindless LLC accounting....
So - me, knowing me, my quirks, how I like to talk to others, what I know and what I find curious and interesting - but infinitely faster with more compute. Maybe she’s one step ahead of me, a little smarter, more ambitious, and just a little “more” where I’m not.
I came across the concept of digital twinning in a Wired article titled “Welcome to Mirror World” by Kevin Kelly in 2019.
The piece explored how we were going to be able to create and interact with a complete virtual twin of our existing world and all the applications for social, manufacturing, autonomous robots, cars, Siri made real, engineering, etc.
This magazine moved with me from Santa Monica to San Francisco and back to Los Angeles.
I held onto this because it gave me hope and inspiration - breathless wonder and a wild flurry of ideas for how our world, duplicated, would intersect the physical and virtual planes.
Kelly’s predictions didn’t include duplicating ourselves, instead focusing on how we may interact with this dimension and with each other bouncing reflections back and forth - a mirror.
Fast forward to 2025, I’m much more interested in what or who my digital twin would be, looking out at me from that mirror.
AI as an Extension
Several panels ago, a speaker threw out an idea about how AI is an extension of ourselves - I drew a little stick figure diagram to simmer on the idea for a while.
The moderator had raised the growing concern around candidates (real or fake) with embellished resumes, armed with a growing suite of AI tools, are circumnavigating recruiters’ ability to effectively screen or evaluate talent.
The panelist started with, “Well, recruiters have treated candidates so poorly for so many years this is what we deserve…” (wrong) but later posited that if a recruiter is using an AI derived job description, and an AI filter to narrow down the pile of resumes…is that not an extension of the recruiter in the same way that a candidate has an AI written resume as an extension of themselves?
And if these two extensions interact or evaluate one another - how is that any different than what we do in a fully human driven interview process?
My knee jerk reaction is that this isn’t a fair comparison of the inputs and outputs of these “extensions.”
Some of the most talented people I know don’t understand how to write a good resume or what one might look like. So they provide poor inputs into an AI and have no way to judge the quality of the outputs. The resume becomes in many cases a distorted reflection (intentionally or not).
The difference between question and answer vs input and output for interacting with any of the AI chatbots is subtle, but consequential.
A recruiter understands the inputs and outputs for a job description, and what to search for in order to find a candidate profile that aligns to those skills and qualifications in a stack of a thousand resumes.
In this example, our twin or mirror is of our own experience - for the candidate (literally) it’s based on their experience and for the recruiter it’s a set of criteria they’ve validated from the hiring team and have set to filter through applicants.
The 3-legged Race to AI Replacement
This breaks down in the next stage of the recruiting funnel - and this is where many talent AI or tech platforms are missing the mark. At the next step you are no longer mirrored, but an extension in which the mimicry goes beyond your knowledge and decision making. We skip over the digital twin who understands the nuance and go straight to a replacement.
When we build solutions to extend the digital self beyond the SME’s knowledge and context, and combine it with AGENCY your physical self is no longer needed.
However, current solutions lack the context and agency - many developers are just looking to replace current processes. Rip out the human, insert AI vs. building from the digital twin outwards.
Let’s extend my sketch:
If we keep replacing each part of a hiring process instead of finding those extensions, we land on not needing humans on either side of the table quickly.
On the candidate side we’re already seeing this manifest - candidates (again, real or fake) are using AI to pass through technical interviews, land the job, and are found out as frauds. In this case a manager confronted an underperforming new hire, and asked him to keep his hands in the frame of the camera to be assured they weren’t getting an AI assisted answer.
And we’re back to finding very “human” ways of fixing this, as predicted, with in person interviews, whiteboard sessions, and referrals.
Let’s assume the growing percentage of candidates that leverage the AI enabled approach for interviewing are currently in a role where AI usage is encouraged, or even required.* Jumping from their AI supported work day to an AI supported interview to land you a job where you’ll be required to use AI to get your job done…feels natural.
*Two things stood out to me in Shopify CEO Tobi Lutke’s recent memo: first, that AI usage would be tied to performance and peer reviews - which in turn logically should show up in interview questions as these two points in evaluating talent must align with one another; second, before asking for headcount hiring managers must prove the work can’t be done by AI.
If your company has implemented AI driven interviews (at least for the recruiter screening stage) your recruiter headcount was slashed or frozen and the remaining team carries double or triple the workload, spending the day approving or declining the AI’s proposed course of action for each candidate (in turn training the AI)…but how long before the AI is making that decision?
And now we’ve arrived where on the other side, AI can just do the job of that candidate, AI can do the job of the recruiter or interviewer - so AI is interviewing AI (this is where all the mirrors and twins smash together like a kaleidoscope). We’ve hit AGI or developed enough specialized agents that we don't need to interview anyone, or the few people left to be interviewed are doing so without any human interaction.
Back to the digital twin, an extension of the known. Can you personally engineer enough of a semblance of a twin to find a use or value add today? This is where platforms are going now with rolling personalization and memory into chats while users are trying to find optimization for themselves.
“The mirrorworld will be the badly needed interface where we meet AIs, which otherwise are abstract spirits in the cloud,” Kelly predicted in 2019.
We aren’t the technician on the factory floor with AR googles on playing with a 3D model of a robotic arm to troubleshoot a pressing issue hidden deep within gears. We’re trying to mesh our flawed human selves with something virtual and superhuman to improve - to accelerate. And first - we have to try and tell it everything we know, want, feel, think - while it simultaneously and hungrily consumes the wealth of all virtual knowledge behind the scenes to answer us better and faster.
We’re holding ourselves up to the mirror of AI to see what we’re missing, to get instructions on how to fill those gaps. Slowly, we’re building our digital twins by attempting to make ourselves more whole.
But once the digital twin surpasses us and our known inputs, there’s a part of us that’s obsolete. Maybe the career part for knowledge workers. When the last SME is no longer needed, what then.
What then?
Work in factories that haven’t been built?
The strawberry fields?
Everyone becomes a CEO?
How does capital work now?
Collectives?
I leapt from Andii2 freeing up my time and brain processing so I could find a new startup to build, write and paint with more hours in my day to another version of the near future working in a strawberry field up the coast.
It’s easy to quickly spin out into Universal Basic Income proposals as we’ll all be out of work while a select few manage our new AGI superpowers.
As technology consistently spasms and leaps forward we hear the refrain [it] will put us all out of work any moment over and over.
Two Worlds
There are two worlds in tech right now on a course to collide. One ignoring, and the other embracing the far reaching cycle of change that we’re on the precipice of.
World 1: Dead walking. Corporate bloat ouroboros monster eating itself doing the same thing over and over again.
World 2: Ruthless efficiency. Adopt and adapt. You may have years but you don’t have another decade to adjust. Move.
The veil between these two worlds is thin and held together by…
Bureaucracy. #1 priority in an organism steeped in bureaucracy is to protect itself at all costs. It’s baked in even after layoffs, developed layers of systems to insulate itself, and nearly impossible to eradicate.
Willful ignorance and unwillingness to leave behind the costly follies of virtue signaling on behalf of executives (ex. talent management functions, DEI money pits, fluffy employer branding, feelings over outcomes).
A distinct lack of “assholes” according to DHH or a cadre of radicals via the “Great Awokening” according to Marc Andreessen (probably both).
Risk aversion to new technology - the legions of full time hires who won’t experiment with anything new for fear their roles really are and become replaceable.
Resistance to accepting the rapidly shortening and intensifying cycles of change - less jobs, more instability, cusp of massive upheaval.
Finally, and where I want to dig in, the tech being developed isn’t close enough to the individual.
These worlds also represent two approaches to building new tech. In Dead Walking, you continue to revert to the mean with software that aims to become too big to fail. In Ruthless Efficiency, we’re still figuring out how we want to build.
Software Reverts to the Mean
Reverting to the Mean: Refers to the statistical phenomenon where extreme or unusual outcomes are likely to be followed by more typical or average ones. In simpler terms, things tend to return to normal over time.
Saas needs to undergo its own rapid breakdown and transformation, traditional Saas cycles (in general) are no longer going to work.
If you’ve spent anytime “in the trenches” - aka a B2B Saas company - you’ll recognize this pattern:
Solve a problem with a small featureset, find product market fit (PMF) with small medium businesses (SMB)
Take money from Venture Capital (VC)
You need reliable annual recurring revenue (ARR) so now you need bigger customers
Start building more features, but promise ALL the features in order to win the mid-market segment (this is acquire customers at all costs time)
Take more money from VCs
Now you have to get the Enterprise customers and these logos require BIG promises
It’s make or break time: SMB is on autopilot, mid-market are complaining but you keep those shiny promises coming, and now your engineering team is pulling all nighters to make THE FEATURE that will win you your first Enterprise customer
Spoiler: THE FEATURE is what one buyer at that one Enterprise logo wanted and this feature will be wholly irrelevant later
Take even more money from VCsStratify your offerings and featuresets into product tiers, try to keep everyone happy
Inexplicably this is around the time you nuke your Customer Success team and try to roll out some half hearted AI features
Learn Enterprise doesn’t want risk, they want their old software but with a better UI and they’ll only switch if their procurement team can confirm you have the same features as the last platform
Now you’re building a slightly better incumbent
The next generation of software comes along and the cycle repeats, but while they scale you (the decision maker) move to a new company and need to select a new software platform but you ask yourself “Why take the risk?”
I’ve already been burned on undelivered feature promises or lose features I’m priced out of in my new “tier”
I’m used to the “mean” software now and I don’t want to have to go through this all over again
Traditional Saas development is a race to the mean.
In 2025 as a software founder you’re trying to meet the needs of companies that aren’t scaling headcount by volume, they’re scaling headcount by capacity and augmenting with AI - and you want some of that budget.
Today, there are 3 approaches to development outside of the Saas feature sprawl:
The rip and replace - wholesale replacement of people or process steps that should still include humans
Tools we didn’t need - building products in search of a solution
Augment and twin - mirroring and user as a true extension of self
However, these tools are still on the edges: add-ons, integrations, or many fragments of a tech stack that you’d need to stitch together. Most founders I’ve worked with have (rightly so) decided to not try and build the next ATS, but to build on top of what most of their customers currently use.
To better understand what these new solutions are complementing, and why most have chosen not to innovate themselves into an all in one solution yet, we need to look at the load bearing platforms.
The Platforms
Two platforms critical for organizations who want to hire, manage, and pay people: the Human Resources Information System (HRIS) and Applicant Tracking System (ATS). There are issues with each market leader - one exemplifying the world of the Dead Walking, another stuck in the cycle of Saas development from a previous era.
Workday: The Bureau of Bureaucracy
Inevitably when the company I’m scaling tells me it’s time for Workday I say two things:
We won’t consider their ATS - myself and this entire recruiting department will walk if you try to force this
We aren’t integrating headcount management, I’ll integrate Greenhouse on the backend to push our hires over
(a lie I tell myself) This is the last time I’m doing this
If there ever was a piece of software that embodied the bureaucracy - how it multiplies and protects itself in ever obtuse layers of confusion - or really a piece of software that gives purgatory vibes…it’s Workday.
It’s the opposite of intuitive. People will get this cult-ish glaze over their eyes and tell you it’s an incredibly powerful piece of software. An entire cottage industry of consulting firms and implementation specialists has been erected in ever widening concentric circles around this software to repeat and profit from this message.
Once you get through the failure of implementing it…it’s too big to fail.
It’s a blank box, they’ll tell you. You can customize it however you want. I say this with love in my heart…HR Operations folks roped into leading these implementations are not Product Managers. So without a Product Manager doing the magic of transforming user requirements, current and future expected behaviors, understanding the long term vision of the organization - it is and always will be a dumpster fire.
It really is a giant, powerful, blank piece of software that once you start building the simplest of workflows you go from 3 clicks to approve PTO to 37 clicks. You can’t do the implementation yourself. You have to bake in (easily) a 6 figure consulting fee to one of the firms that Workday has endorsed and please believe me when I say - ANY change, discovery, or new input into how you want this configured they will tack on additional fees, push out your go-live, and fight you tooth and nail (you may be the customer but in Workday land you are never right).
You have to hire at minimum one full time person just to admin the thing once you go-live. The bureaucracy multiplies. Strength in numbers. Need more people to keep the software running. Gotta integrate. Finance is the next target.
The only light at the end of this tunnel is my sincere hope that Rippling eats their lunch. That is of course after they’re done extraditing Deel executives from Dubai.
Even then, there will be those in the Enterprise that will never switch. The difference in switching costs for HRIS vs ATS platforms is enormous. The pain you went through to make Workday function (such as it is) will never be forgotten…and now you have a team of Workday admins and specialists and their jobs are solely to keep this system up. Bureaucracy established.
Greenhouse: Reverted to the Mean
Greenhouse as a tool meant for recruiters. While the UX is a little disorienting to hiring managers who generally tend to prefer Lever, it’s a workhorse. Integrates with everything. Gets the job done.
Where did they go wrong?
First, they did everything right by doubling down early ensuring they could integrate with every part of the talent and HR tech stack - and fast. (This wasn’t table stakes in 2014-2015 and set them apart in the space). But over time, this became a crutch, keeping them from really becoming a true platform.
Scheduling didn’t advance, so I have to pay for something like GoodTime.
Analytics were easy for me (a spreadsheet nerd) but I needed to stack on TalentWall for everyone else.
Then came Gem (fka ZenSourcer) for a CRM, and and and…
Greenhouse was at the hub but it wasn’t the full stack and wasn’t bleeding out to the edges to be my twin.
TalentWall came on the scene in 2016 for analytics - recruiters, recruiting managers, hiring managers, C-Suite execs, chiefs of staff…everyone had an accessible way to see exactly what was happening with one search or their entire team’s hiring.
TalentWall had hacked a version of me armed with spreadsheets, spending my evenings building out power points and graphs, developing like a digital twin.
But instead of acquiring them (Crosschq did in 2022) Greenhouse slowly…slowly built out versions of the “wall” without any of the ease of navigation or intuitive features.
Building Like a Digital Twin
Building through the lens of a digital twin requires you to understand the user in ways that you’ve never had to really grasp before. This isn’t “user wants feature x so I’ll make that” it’s “how does user move through their day?” and “why are they working late” or “how does the user think?”.
This next generation of software is that it doesn’t rely on feature sprawl or building for the Enterprise - it’s tailored to the individual. As we scale down to smaller and smaller teams that are increasingly efficient, it’s about the individual user. If this user could duplicate themselves - what would the twin do?
These are just 3 of the arduous, admin-heavy problems where we need an augmented version of ourselves to extend our own efficiency in talent…
Negotiating Candidate Profiles
Successful hiring begins and ends at building a job description. Recruiters pull from their experience, market research, compensation tools, and on and on to validate and make arguments for things like…
…why you won’t find a founding front end engineer for an onsite role in NYC who’s also really good with customers for $135k. (And how we can adjust those requirements so we do find the “right” person.)
It’s time consuming to pull this together, recruiters may miss something or not have access to a critical data point like market compensation for the specific role, it may be outside their area of expertise, etc.
The first push back from the hiring manager is re: the recruiter’s assessment of what needs to change, second push back demands the source data, the third usually calls me in to escalate. It’s time intensive and can damage new or fragile relationships between teams quickly. Now we’re now arguing the merits of the proposed candidate profile vs balancing wants and needs against the market.
But what if - the requirements (good, bad, and ugly) went into an AI tool that told you - as written here’s your candidate pool, or here’s the dealbreaker requirements that are “wrong” and the data to back it up, and by the way I’ve sourced a few profiles and put together some candidate archetypes or personas, validated compensation, and found several companies with similar tech stacks to target.
Here, the AI is filling in the gaps of a recruiter’s knowledge (wherever those may be), providing valuable insights (fast) that allow the recruiter to go back to the hiring manager to make the case on why the profile should look a little more like…this. With less lift, less bias, and more detail.
Now imagine if you could evolve those finalized job requirements into…interview scorecards, a 30-60-90 day plan, performance review questions…
Scheduling
We haven’t seen a tool yet that can handle complex panels, multiple timezones, and has removed the need for a full time human to handle candidate scheduling.
You’re not just solving for availability and time slots and Zoom provisioning - the human intervention is knowing the preferences of the interviewers. Interviewers don’t update calendars, you have to remember that one executive has a light hold on certain days except for candidates from this one search they’ll prioritize over dinner time or a pipeline review meeting (but only the one on Thursdays).
The scheduler needs to also handle rescheduling, new conflicts, and ideally would provide guidance to ensure the bulk of candidates make it through one interview stage within an ideal time frame to avoid recency bias for interviewers. Plus escalate how scheduling logistics are impacting search timelines overall with suggestions for improvements.
Let me break up with LinkedIn
LinkedIn - the walled garden we love to hate and hate to buy access into - has somehow become even more difficult to deal with. You can no longer purchase a la carte Recruiter seats - you have to also buy job slots, and get pushed into paid company pages as well…for one Recruiter seat you’re now looking at $20-30k/year. Free job slots are falling off their offerings.
The mafioso vibes are real: You could get a trial or you could…just make it easy on yourself and get locked into a 3 year deal.
The person who can develop NOT the “next” LinkedIn but the search tool that divorces us from having to pay the LinkedIn admission price…is going to strike gold. They’ll have to overcome 3 things first:
You’re not replacing one database with another - you’re giving us a search tool that gets us the contact information and up to date digital profiles of candidates from everywhere
You’re not going to re-create LinkedIn pricing - again because the tool you’re selling isn’t the data, it’s the ability to search
You’ll have to prove that you can match or beat the results for your platform vs. LinkedIn
Easier said than done
The space is dominated by companies trying to skip over these ideas in a race for an AI recruiter - any CEO’s dream: I give it my demands, it gives me people. Others, realizing their buyers for the moment are the TA teams they aim to replace (or the HR teams that see themselves as next up on the chopping block) are taking a more sly approach.
We’ll take all the administrative work off your plate so you can focus on strategy…
Rip and Replace
The lie starts out easy enough - you don’t want to do admin work, you want to be strategic, a partner to the business…
Side note: Stop saying you’re a business partner or the hiring managers are your clients. You’re part of the business. You are the business - you’re the one building the company. This is just another version of the whiny hand wringing from 10 years ago about how “we need a seat at the table” - you built the table and hired everyone else around it - take a seat. Stop asking permission for the things that are already yours.
Back to the lie - a lot of our organizations (and software developers) don’t know what the “admin” work is - so they replace the first obvious thing: note taking. Now we have a flurry of note taking apps that hang out in every meeting, and AI note takers have become a baseline feature for talent solutions.
Is it truly note taking? You’re a transcription with a summary that can be hit and miss, often needing a rewrite. Now you’re admin-ing the admin tool that was supposed to make this easier. Arguably, taking notes in an interview or writing your own observations while using your own critical thinking is a really important part of being a recruiter (or any profession). Querying the transcript to see if this candidate demonstrates “grit”…isn’t it.
The practice of interviewers documenting their own observations (bringing in the necessary context and agency) after the interview ends won’t happen when the alternative is clicking “generate with AI” to summarize scorecard notes. Why have a hiring manager on the call at all? AI tools often automate the wrong thing.
But there’s something more insidious in the line of admin vs strategic.
It assumes that the people signing off on budget believe TA is strategic and that we have strategic work to do.
If that’s in question, there’s a serious failure in talent leadership - however, it’s a case that must be argued and proven repeatedly to the business to avoid backsliding from the #1 revenue generating team to a cost center.
Back to Tobi’s memo - if you needed another tech recruiter to execute on a ballooning headcount plan, you now need to justify why. Why can’t this be done with AI? So your team remains small, your tech budget makes up for the people you’d typically hire, and your scope narrows to tech, efficiency, and capacity (much more so than any other point in your career).
But if those AI tools you implement are less of a digital twin and more of a forced replacement when the process still needs to be rooted in human agency and context, and your workload is now (by design) orders of magnitude larger - how are you able to do anything strategically?
Collapse to the Collective
There’s a collapse of all things down to the individual. One person plus AI can make a company. To make good AI, you have to mirror the individual. At the same time, the burden on the individual to develop and fuel their own agency has never been more urgent.
Can you build something?
Can you figure out something new?
Can you design and combine your digital twin?
How can you scale yourself?
Right now, AI isn’t worthwhile without a subject matter expert (SME) to provide strong inputs and evaluate the outputs. Each week, someone else becomes the main character on X for vibe coding something with a massive security issue. You don’t know what you don’t know.
Currently, without a SME you’re running the real risk of whatever you vibe [code, market, sell, recruit] breaking, being dangerous, or having a real meltdown. But soon (now-ish), you’ll see small collectives of SME’s, augmented with AI, building long term, sustainable, successful businesses. I think this is different from a company so I’ll call these collectives.
We evolved from company to “we’re a family” to team to organization. And now in the rise of the individual which I believe could quickly evolve into something new: a collective. You contribute to the whole, but retain enough independence to leverage your expertise across multiple projects, ideas, and experiments. A small, trusted network of SME’s working together.
And maybe your collective isn’t vulnerable to the whims of a cult of personality, virtue signaling, or politics. It’s democratic, profits are shared equitably, and you have the space and freedom to develop your own interests and passions untethered from a sole full time make or break career. And maybe your digital twin is right there with you.
Until then, happy building. And welcome to mirrorworld.
If you’re developing talent tech and you’d like a second pair of eyes, an advisor, or just a user feedback session, I’d love to chat. If you’d like to keep up with the blog, subscribe below and follow me on LinkedIn for more takes.
