Loading…
AI Agent Development · Dallas
Provisioning ops at an AT&T-scale carrier in the metroplex still chew through tens of thousands of interconnect tickets a quarter. LERG lookups, LOA validation, ASR keying, network inventory checks, FOC date chasing. Each ticket is a 60 to 90 minute manual chain across four systems that almost talk to each other. An agent that owns the deterministic path closes most tickets before a senior engineer sees them.
The same shape repeats across the rest of Dallas. ExxonMobil pipeline-integrity ops drafting PHMSA-7100 narratives against fixed 30 and 60 day windows. Southwest IRROPS desks rebooking tens of thousands of passengers after a single weather event. Tenet and Baylor Scott & White claims teams running Chapter 542 prompt-payment workflows at six-figure monthly volume. Every one of these is an agent-shaped problem.
Every engagement is fixed-scope. Discovery first, fixed quote after, four to eight weeks of build with weekly demos on Central time.
Describe the workflow.
Dallas concentrates the kind of back-office work that an agent handles well: high-volume, rule-bound, multi-system, with a regulator or a service-level commitment on the back end. Telecom provisioning at AT&T Mobility and the wholesale carriers in Richardson is the first shape. A wholesale interconnect order bounces through the order-entry tool, the LERG, the inventory database, the design tool, and the field-dispatch system before a single port is lit. Engineers spend more time chasing data across screens than designing the circuit.
Energy ops at ExxonMobil in Irving, Pioneer Natural Resources, and the midstream operators run a different version of the same problem. PHMSA pipeline-integrity reporting under 49 CFR Part 192 and 195 has fixed reporting windows that do not bend. The data lives in OSIsoft PI, in the inspection database, in ArcGIS layers, and in the SAP work-order system. Assembling a single Form PHMSA-7100 narrative is a half-day job for a senior engineer. The Texas Railroad Commission permit workflows have the same structure.
Airline ops at Southwest and American Airlines is the third shape. IRROPS exception handling during a storm is mostly deterministic rebooking against published rules, with a smaller set of complex cases (multi-segment, ADA, unaccompanied minors, groups) that need human judgment. Headcount at the IRROPS desk does not scale with weather. An agent that handles the deterministic 65% lets the desk focus on the complex 35%.
Healthcare claims at Tenet, Baylor Scott & White, and the regional payers in Plano is the fourth variation. Texas Insurance Code Chapter 542 imposes clean-claim and prompt-payment rules with TDI documentation requirements. The workflow combines 270/271 eligibility lookups, NPI and CPT validation, prior-authorization status checks, and denial-letter drafting. Claims processors spend most of their day moving the same fields between Epic, the clearinghouse, and the payer portal.
Frito-Lay supply-chain ops at the Plano HQ run a fifth variation around vendor onboarding, route exception handling, and trade promotion ops. Match Group runs trust-and-safety triage at scale in Uptown. Jacobs Engineering coordinates infrastructure project intake across hundreds of public-sector RFPs. Different verticals, same operational shape.
Every one of these workflows is an agent candidate. Not a chatbot or a Power Automate flow. An agent with tool use, structured output, a regulator-grade audit trail, and a human gate at the irreversible action.
Six components show up in almost every agent we ship for telecom, energy, airline, and healthcare ops in the metroplex.
LangGraph state machines for plan-execute or ReAct flows. Transitions are explicit and reviewable, not buried in a prompt. The graph is testable in CI with a labeled trajectory set before it touches production traffic.
ServiceNow, SAP, Epic, OSIsoft PI, ArcGIS, LERG feeds, AT&T provisioning APIs, Sabre, Southwest's internal IRROPS interfaces, TDI 270/271 transactions. Every tool call has a typed schema, retry policy, and a fallback for degraded upstream.
Short-term context lives in the agent state. Long-term memory uses pgvector or Pinecone, indexed by ticket number, claim ID, NPA/NXX, asset tag, or whatever your row-level key is. The model does not invent context.
Any irreversible action (book a rebooking, submit a PHMSA form, deny a claim, push a provisioning order) routes to a human approver in Slack, Teams, or your existing queue. The agent waits with the full reasoning trace attached.
PHMSA inspections, FCC audits, TDI reviews, DOT customer-service plan checks. Every prompt, tool call, decision, and override is written to an append-only log with model and prompt versioning so any past decision is reproducible.
Task completion rate, hallucination rate, and tool-call accuracy measured against a labeled set you provide before the agent goes near production. Regression suite runs on every prompt change in CI.
Dallas-Fort Worth is the largest metro in Texas and the second-largest concentration of corporate HQs in the country. The shape of the back office here skews toward high-volume regulated ops: AT&T runs its network operations from Dallas, ExxonMobil runs upstream and midstream from Irving, Southwest Airlines runs its entire dispatch and IRROPS operation from Love Field, Frito-Lay runs North American supply chain from Plano, and Tenet and Baylor Scott & White run the largest hospital networks in Texas. Texas Instruments and Match Group fill in the tech side.
The unit economics of an agent are unusually clear in these verticals because the work is high-volume and the alternative is hiring more analysts. A 30% reduction in carrier-interconnect ticket handle time at a wholesale carrier saves several million dollars of provisioning-engineer time annually. A 40% cut in claims processing time at a regional payer absorbs growth without adding headcount.
We work remotely with Dallas clients on Central time. Discovery calls fit between 8am and 6pm CT, demos happen weekly, and the codebase ships to a repository you own. No agency-style offshore handoffs, no overnight blocking dependencies. The team overlaps with your business hours by design.
Industries where we see strongest fit: telecom carrier provisioning, pipeline and upstream energy ops, airline IRROPS and dispatch, healthcare claims processing, consumer-packaged-goods supply chain, and engineering-services proposal ops.
Tell us what your ops team does every day that you think a software agent could handle. We'll reply within one business day with a rough scope and a price range. No commitment, no sales call required.