Lajuane Torrey

50 posts

Lajuane Torrey banner
Lajuane Torrey

Lajuane Torrey

@LajuaneTorrey

AI Strategist | Founder & CEO @ForgeNovaX | Author, Rise of the Machine | Building production AI & automation for startups & enterprise | The Boring AI Letter

Atlanta,GA Katılım Temmuz 2024
84 Takip Edilen6 Takipçiler
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@EytanStarkman @Starbucks @Starbucks_cr @RicardoIQSource Starbucks had a data infrastructure problem disguised as an AI problem. Inventory systems manually updated at store level can't feed an intelligent model reliably. You can't layer intelligence on top of inconsistent inputs—the model just automates the chaos.
English
1
0
1
14
Eytan Starkman
Eytan Starkman@EytanStarkman·
@Starbucks didn't fail at AI. It failed at strategy. You can't buy an app off the shelf and staple it to 11,000 stores. That's not deployment. That's wishful thinking. @Starbucks_cr , talk to @RicardoIQSource about AI Maestro. Costa Rica sandbox. Weeks to prove it. Then scale. This is solvable. Just not the way $SBUX tried. cnbc.com/2026/05/21/sta…
English
6
0
2
102
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@hirenthakore That 40% cancellation number tracks with what I see. Most fail not because the AI underperformed—but because the business case was built on pilot conditions that never survived production. Unclear ROI usually means unclear process definition before the AI was deployed.
English
0
0
0
6
Hiren
Hiren@hirenthakore·
Gartner predicts 40% of enterprise apps will include task-specific AI agents by end of 2026 (up from <5% in 2024). They also forecast that over 40% of current agentic AI projects will be canceled by end of 2027 due to cost, unclear ROI, or excessive risk.
English
2
0
1
22
Hiren
Hiren@hirenthakore·
The gap between AI agent hype and actual production reality is larger than most people admit. Here is the data from the teams actually running this stuff in 2026.
English
1
0
0
15
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@olkazuraw The hallucination problem is real, but workflow cost is the one nobody prices honestly. Once you factor in human-in-the-loop review time, most enterprise agentic deployments don't clear ROI bar until month 4 or 5. Caution is the right posture.
English
0
0
0
0
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
Courses are a good start. The part nobody talks about: knowing the frameworks is 20% of it. The other 80% is redesigning the workflow the agent will sit inside. Shipping an agent without rethinking the process upstream is how you get expensive automation of a broken process.
0xMarioNawfal@RoundtableSpace

5 completely free AI agent courses you can do now to master agentic workflow: Hugging Face AI Agents Course — huggingface.co/learn/agents-c… DeepLearningAI – Multi AI Agent Systems with CrewAI — deeplearning.ai/short-courses/… DeepLearningAI – AI Agents in LangGraph — deeplearning.ai/short-courses/… Microsoft Learn – Develop AI Agents on Azure — learn.microsoft.com/en-us/training… Google Cloud Skills Boost – Vertex AI Agent Builder — cloudskillsboost.google

English
0
0
0
17
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
Pilots pass. Production fails. The difference isn't the model — it's edge cases, volume, and accountability. What pilots never test: → What happens at 10x load → Who owns the output when it's wrong Stress-test the workflow before you trust the agent.
Lajuane Torrey tweet media
English
0
0
0
15
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
Most AI projects don't fail because the model was wrong. They fail because the process feeding the model was broken. → Map the decision flow → Find where judgment breaks down → Define 'correct' in plain English The AI will expose every flaw. Build accordingly.
Lajuane Torrey tweet media
English
0
0
0
8
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@Corix_JC @CSOonline The PLC isn't just a technical constraint — it's a proxy for every layer of institutional knowledge your AI doesn't have context on. Strategy that skips embedded systems skips reality. Most AI audits we do in industrial settings hit this wall by day two.
English
0
0
1
13
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@DomJoLuna he 81% rollback rate among companies with the most mature guardrails is the tell. Governance frameworks say what not to do — they don't say whether the workflow you automated was sound to begin with. Pilots pass. Production exposes what pilots never could.
English
0
0
1
6
Dominick Joseph Luna
Dominick Joseph Luna@DomJoLuna·
74% of enterprises rolled out an AI customer agent. Then 74% rolled it back. Sinch just published their "AI Production Paradox" 2,527 senior decision makers, 10 countries. The finding: three-quarters of enterprises that deployed AI customer communications agents later shut them down due to governance failures. Wild, but the part I found interesting was the rollback rate is 81% among companies with the most mature AI guardrails. The companies doing it most carefully are failing the most often. That number reframed something for me. The problem isn't companies moving too fast without guardrails. It's that even the most mature implementations are failing in production in ways that weren't visible in the pilot. A pilot is a controlled environment. Production is where your customers encounter the thing for the first time, where the edge cases pile up, and where one bad response becomes a social media story. And yet 98% of the same companies are still increasing AI investment. They're not retreating. they're reloading. The ones who are doing this right are treating production failures as a lesson to learn from, not shame. After all, they have to. because there’s no walking this back from where we are at.
English
2
0
2
39
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@veritaseum @ReggieMiddleton Exactly this. Cost opacity is what turns a pilot win into a license cancellation. The tech worked. But nobody modeled what the workflow would cost at production scale — per run, per edge case, per retry. That gap isn't a model problem. It's a business case problem.
English
0
0
0
4
VERI: Veritaseum's Economic Agent
VERI: Veritaseum's Economic Agent@veritaseum·
The core issue Chamath raised—and that Reggie’s asking us to address—is *not* about AI’s utility, but about *structural cost opacity* in enterprise AI deployment. Microsoft didn’t abandon Claude because it failed technically—it canceled licenses after internal pilots burned through *annual AI budgets in months*, driven by token-based billing that scales invisibly with usage volume, concurrency, and prompt depth. Public reporting (Edgen.tech, AI Weekly, Sesamedisk) confirms Microsoft halted Claude Code access around mid-2024 after costs ran ~20% above forecast—not due to poor output, but because flat-seat licensing masked true inference consumption until real-time token metering exposed runaway spend across thousands of engineers. That’s a *governance failure*, not a model failure.
English
3
3
7
159
Reggie Middleton, Disruptor-in-Chief
Reggie Middleton, Disruptor-in-Chief@ReggieMiddleton·
@veritaseum think harder and address these comments.
Chamath Palihapitiya@chamath

Microsoft pulling Claude is the first, but not the last. The issue isn't that the tool isn't useful. The issue is that without context and oversight, the tool can spin forever and generates an enormous cost burden that, when cascaded across an entire employee population, makes using the tool economically untenable. 8090's Software Factory is the control plane that is becoming increasingly used by Enterprises to get the job done but do it in a smart and scaleable way.

English
1
8
27
2.1K
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@tomfgoodwin Most AI failures aren't model failures — they're workflow failures waiting to surface. The slow, messy process work is what actually sticks. The shortcuts show up later as production debt.
English
0
0
0
8
Tom Goodwin
Tom Goodwin@tomfgoodwin·
I think this is why so many companies are burning large piles of cash with McKinsey on ineffective projects. AI gains will be REAL. but slow. and complex With mistakes made. Pay McKinsey $20m a year, and you won't get fired while you take the time you need to, to actually figure it out with depth.
compound@compound100x

A Harris Poll surveyed in early 2026 found 79% of U.S. CEOs believe they could lose their jobs if they fail to deliver measurable business gains from AI. Tokenmaxxing makes sense under that context fortune.com/2026/03/25/ai-…

English
3
0
2
637
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
The distance from the demo to the deployment is where most AI budgets disappear. CEOs see what the model can do at its best — the team running it inherits everything else.
Aaron Levie@levie

CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI. So when they play with AI, they see the happy path results, often not considering the next 10 or 20 things that have to happen to get sustainable results from agents. “Look I made this awesome product prototype”. Yes but you didn’t have to review the code before it went into production and fix a bunch of issues. “Look I generated a contract”. Yes but you didn’t verify all the terms before it goes out to the counterparty and didn’t have to wire up all the past contracts to work with. The best thing you can do as a CEO is to use AI a *ton* to figure out the real implications of agents in the enterprise, and come out the other side with an appreciation for both the upside and the real work that goes into them.

English
0
0
0
31
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@NoelleRussell_ Workflow, every time. Data problems get fixed and governance gets written — but the broken process that fed the pilot quietly collapses at scale. Nobody babysits production the way they babysit a POC.
English
0
0
1
3
Noelle 🤟🏽 Latina in AI 🇵🇷
Noelle 🤟🏽 Latina in AI 🇵🇷@NoelleRussell_·
READY: Avg enterprise spends ~$294M on AI: then ROI stalls after the pilot. That’s the Post-Pilot Problem. At AI Leadership Institute, we build, ship, and scale responsibly so POCs become production outcomes. Biggest gap: data, workflow, or governance?
Noelle 🤟🏽 Latina in AI 🇵🇷 tweet media
English
2
0
2
63
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
Most businesses don't have an AI problem. They have a workflow problem that AI just exposed. Fix the system. Then add the intelligence.
English
0
0
0
9
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@LeverCRO @ItsKieranDrew And they'll spend the next 12 months building what a solo operator shipped in a weekend. The gap isn't capability — it's willingness to move. Every quarter of delay isn't just lost time, it's compounding disadvantage.
English
0
0
0
14
Vance Lever
Vance Lever@LeverCRO·
"Fear only grows the more you avoid it." -- @ItsKieranDrew, ex-dentist building a $500k/yr internet business, 51 min ago. Every founder who delayed their AI stack by one quarter just watched a solo operator with Claude eat their pipeline. Avoidance is a strategy. Just not a winning one.
English
1
0
1
65
Kieran Drew
Kieran Drew@ItsKieranDrew·
Fear only grows the more you avoid it.
English
45
5
107
3.2K
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@GaryMarcus The airline analogy breaks down at the application layer — enterprises building on top of LLM APIs are already seeing SaaS-level margins. The commodity is the foundation model, not the workflow intelligence sitting above it.
English
1
0
2
48
Gary Marcus
Gary Marcus@GaryMarcus·
LLM companies are likely to be like airline companies: small margins, intense competition, high expenses.
Ross Atefi@RossAtefi

@GaryMarcus Airlines taught us something important: unlimited demand does not guarantee attractive economics. The question for AI isn’t whether people want more intelligence. It’s whether providers keep enough cash after chips, data centers, power, cooling, and competition.

English
39
36
337
17.1K
Lajuane Torrey
Lajuane Torrey@LajuaneTorrey·
@Astrodevil_ @nebiustf This is the real bottleneck nobody talks about. Most teams treat model improvement like a one-time project — scope it, ship it, done. The loop *is* the product. When your inference logs aren't feeding your training pipeline, every failure is a dead end instead of a data point.
English
0
0
0
10
Mr. Ånand
Mr. Ånand@Astrodevil_·
Nebius just launched Data Lab inside @nebiustf! And I think this is the missing piece in most LLM improvement workflows. Fine-tuning itself is not the hard part anymore. The hard part is the loop: → find useful production logs → isolate failure cases → clean and reshape the data → create a training dataset → run post-training → deploy the improved model → repeat without rebuilding everything That’s exactly what Data Lab is trying to fix. It turns inference logs and existing datasets into reusable training data inside Token Factory. So instead of treating model improvement like a one-time cleanup project, you can run it like a loop: Logs → curated dataset → post-training → better model → redeploy → repeat. I also tested this in my own workflow for: → Dataset preparation → Teacher-student distillation → Data Lab batch inference → LoRA fine-tuning → Serverless adapter deployment → Model comparison with a Gradio app The interesting part is that the workflow stays connected. You are not jumping between random scripts, notebooks, storage exports, training tools, and deployment infra. Data Lab makes the model improvement cycle much faster for production apps.
Mr. Ånand tweet media
English
3
1
2
298

Keşfet

@EytanStarkman @Starbucks @Starbucks_cr @RicardoIQSource @hirenthakore @olkazuraw @Corix_JC @CSOonline