Register and share your invite link to earn from video plays and referrals.

Alex Lieberman
@businessbarista
Family first (husband & girl dad) Founder second (@tenex_labs, @morningbrew, @storyarb, @youdistro) Enterprise Applied AI Partner 👇
Joined March 2017
3.7K Following    318.5K Followers
I’m not at all surprised so many enterprise AI deployments fail. So many things have to go right for an agent to make it to production & if one cog in the machine breaks timelines are pushed back (best case scenario) or leadership’s trust is eroded (worst case scenario). Everything that has to work: 1) Scope has to be clear - One workflow with a named business owner, a baseline metric, and a dollar outcome. - Autonomy matched to blast radius: suggest, then act with approval, then act autonomously. Earn each step with data. - Kill criteria defined up front. - Clarity around the current way of working/process and what the desired future state looks like. 2) Data & system access has to configured and trustworthy - Read/write access to systems of record, legacy APIs, and unstructured sources. - Service identities, SSO, and permission models that work for a non-human actor. - Data quality and freshness. Bad source data becomes confident bad output. 3) Evals must be core part of infrastructure - Golden set built from real historical cases, including edge cases and adversarial ones. - Success measured against the human baseline, not against perfection. - Regression suite on every prompt, model, or tool change. 4) Architecture and reliability are sound - Deterministic scaffolding around the LLM. Use code for anything that can be code. - Well-designed tools, timeouts, and fallbacks. - Model abstraction so you can swap providers. -Latency and cost budgets per task, plus rate-limit handling at scale. 5) Security is buttoned up AF - Least-privilege agent identity and scoped tool permissions. - Prompt injection and exfiltration defenses, especially with untrusted inputs (emails, documents, web). - Full audit trail, tenant and data isolation, residency, and vendor DPAs. 6) Governance and risk considered from the jump - IT, legal, compliance, and finance sign-off. - Regulatory fit (SOX, HIPAA, GDPR, EU AI Act, and industry-specific rules). - Human-in-the-loop thresholds, a kill switch, and one accountable owner for agent behavior. 7) Observability and operational rigor - Step-level tracing of every decision and tool call. - Monitoring for cost, quality drift, and failure clusters. - Versioning and rollback for prompts, models, and tools. - A feedback loop from human corrections back into evals. 8) The conditions for adoption are ripe - Executive sponsor with budget authority, plus the frontline process owners involved from the start. - Agent embedded where people already work (Slack, Salesforce, Outlook), not a new destination. - Clear escalation paths and honest handling of job-displacement anxiety. - Incentives aligned so the team benefits from the agent working. 9) Economics underwritten and measured - Fully loaded cost per completed task (inference, infrastructure, review labor, maintenance) against the human cost. - ROI measured on the baseline set before launch, not estimated after. - Understood if it’s part of experimental, efficiency, innovation, or infrastructure budget Until you’ve truly spent time in the guts of an enterprise, you don’t realize just how hard meaningful change is & just how much has to happen beyond raw frontier intelligence.
Show more